Syfert Legal Research MCP server
com.syfert/legal-researchsyfert.comstreamable-httpv2.0.0Scanned
Verify legal citations, case treatment, quotes and whole briefs against 10.7M U.S. opinions
22 tools: 1 can delete or cancel things, 7 change things and 14 only read.
- Reads private data · 0
- Reads text others write · 3
- Can send data out · 0
- 10 findings
Tools listed without sign-in. Remote posture
With Anlyon
1 tool here can delete or cancel things. Govern it.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the Syfert Legal Research MCP server can do
The Syfert Legal Research MCP server is published by syfert.com, a domain the MCP Registry verified. It has 22 tools. It can delete or cancel things (1 tool: brief_delete) and create or change data (7 tools: check_citation, verify_quote, check_brief, suggest_treatment and 3 more). 14 tools only read.
3 tools bring in text other people can write (check_citation, find_issues and get_brief_results).
The fixed checks matched 10 findings (long-description, credentials-mention and annotation-conflict). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in, and its tools changed 3 times since Anlyon Radar first saw them.
Last scanned Nov 20, 2026, version 2.0.0.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
22 tools
Deletes or cancels 1Claude Code: deny
brief_deleteSigned-in accounts only (a free account at https://syfert.com/mcp/; anonymous callers use check_brief with text).1 finding
Why it is in this group
- The name "brief_delete" says "delete": a word for removing or ending something.
- The server marks it not destructive (destructiveHint: false), but an annotation never lowers a class.
Findings
- Warn
The server marks "brief_delete" not destructive (destructiveHint: false), and its own name, description or input give the class "destructive".
What this check looks forMatched text
annotations: {"readOnlyHint":false,"destructiveHint":false,"idempotentHint":false,"openWorldHint":false}; The name "brief_delete" says "delete": a word for removing or ending something.
Description, as the server wrote it Signed-in accounts only (a free account at https://syfert.com/mcp/; anonymous callers use check_brief with text). Delete an uploaded brief and its results from the server now (they are otherwise deleted after the last page is read, or after an hour idle).
Annotations the server declares
- title: Delete an uploaded brief now
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: false
Changes things 7Claude Code: ask
brief_uploadSigned-in accounts only (a free account at https://syfert.com/mcp/; anonymous callers use check_brief with text).1 finding
Why it is in this group
- The name "brief_upload" says "upload": a word for a change.
Findings
- Info
The description mentions "credential".
What this check looks forMatched text
description: "...s are held in server RAM only, for your credential alone, and are deleted right after the..."
Description, as the server wrote it Signed-in accounts only (a free account at https://syfert.com/mcp/; anonymous callers use check_brief with text). Upload a brief of any length for check_brief, in pieces. Send the text in order, about 40,000 characters (10k tokens) per call: the first call returns brief_id; each next call passes brief_id and seq (2, 3, ...); the last passes final=true (a short brief can go in one call with final=true). Then call check_brief with brief_id. The text and its results are held in server RAM only, for your credential alone, and are deleted right after the last page of every section is read, after an hour idle, or by brief_delete.
Annotations the server declares
- title: Upload a long brief for check_brief
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: false
check_briefAudit ALL citations in a draft brief, motion, or memo at once (wraps syfert.com Brief Check).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Findings
- Info
The description is 6,714 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...are and stats.polarity_check_note says so; no finding on those means not checked."
Description, as the server wrote it Audit ALL citations in a draft brief, motion, or memo at once (wraps syfert.com Brief Check). Extracts every reporter citation, verifies each exists, checks the case name the document claims against the real case (hallucination detection), checks each quoted passage word for word against the cited opinion's own text (misquote = not in that opinion; unquoted_by_courts = the text could not be compared and no court has quoted those words from the case: check it with verify_quote), traces an unmatched quote to the case the language actually ORIGINATES in (quote_origin: attribute with "(quoting …)" or cite the source), and returns treatment flags with replacement suggestions for red-flagged authority. Also audits every STATUTE, REGULATION and COURT-RULE citation in the document against the corpus that holds it, reported separately under "statutes" (verified=false means not found in that corpus, which is not proof the section is wrong). ALWAYS run this on a finished draft before delivering it. Use mode=attack on an OPPONENT's brief to get ammunition: distinguishing and criticizing citers for each of their cites. Text over the size cap (128KB anonymous, 512KB with a token) is cut before the audit and the response then carries stats.truncated_bytes plus an INCOMPLETE AUDIT warning. corrections lists every quote the cited opinion does not carry verbatim, with replace_with (the opinion's own words, or, for verdict statute_text / statute_misquote / rule_text, the statute's or rule's own words; on verdict misattributed replace_with is null and cite_instead gives the citation) and star_page, and every unresolved or name-mismatched citation. Each unresolved citation carries verdict (exists_unindexed = a real case our index lacks, confirmed by a citing opinion or by name -- keep it; not_indexed = unknown, weigh volume_coverage (thin 2019+ volumes make a miss weak evidence); interior_mismatch/strong = likely fabricated), so do not report every unresolved cite as a hallucination (MCP13_20260925). A pinpoint page written as a citation of its own is status pin_reference (stats.pin_references, not unresolved) and its correction gives cite_as (first page, then pin). Public-domain cites ("2023 IL App (1st) 220520", "2023-Ohio-1234") are read like reporter cites, and a paragraph pinpoint ("¶ 45", "¶¶ 12-14") is checked against the opinion's paragraph numbers: cites[].para_pin.verdict interior_pin / interior_mismatch, quote_checks[].para_pin.inside says whether a quotation sits in the pinned paragraph; a mismatch is also a correction (MCP17_20260925). Each verified statute finding carries currency (edition, current-through, last amendment where recorded) and statutes.corpus_currency the edition of each corpus: a brief about past conduct needs the text in force on the operative date (check_citation or get_statute with as_of). Pass forum (the court you will file in, e.g. forum="ca9" or forum="fl") to have each unpublished cited case checked against the citation rules of that court: cites[].citability.forum.may_cite is yes / persuasive / conditional / no, or unsure when our verified rules data cannot settle it for that court (it says why: another court system, a district court applying a circuit rule, a pre-2007 federal disposition in another circuit, or a rule that is an open question on the citability review page). An unresolved citation printed in the corpus only by opinions that say it does not exist (sanctions / show-cause orders on fabricated citations) is verdict strong, verdict_evidence cited_as_fabricated, with fabricated_citers. Case rows with a non-green flag carry treatment.flag_reason. A former U.S. Code appendix cite is listed with status former_appendix and the current section. Each pinpoint cite also carries cited_for: a machine-written G-Cite label of what citing courts cite that page for (not any court's words; n_citing is null: no count of courts behind a label is known; a proposition_mismatch / pin_suggestion block adds witness machine_summary and n_citing_page / suggested_n_citing_page = the citing cases in G-Cite's unit for that page). When the brief's sentence fits that page poorly and another page of the same case clearly, verdict proposition_mismatch names suggested_pin_cite: a suggestion to read that page, never a correction to apply unread; the citation itself stays verified. Each case cite with a proposition also carries cited_for_check when the brief's point differs in structure from what citing courts cite the case for: proposition_contradicted (courts cite it for the opposite), proposition_entity_mismatch (a different amendment, statute section, burden or standard), proposition_uncited (nothing like it); confidence hard = a correction, advice = a note. Long answers are shortened to fit the client result cap (size_governor says what was shortened; corrections are always complete); detail="full" returns everything. California cites the Supreme Court depublished or took up on review get cites[].citability (kind ca_depublished / ca_review_granted / ca_transferred) and a CALIFORNIA CITABILITY warning line. A cite whose support_check says available can be laid out next to its sentence with context_detector. cites[].court_check / corrections court_mismatch: the court written with a citation is not the court of the case at it. A brief too long for one call (over ~40,000 characters, more than 200 citations, or over the size cap), signed-in accounts: upload it with brief_upload, then call check_brief with brief_id instead of text; every citation is checked and the answer is paged. proposition_contradicted can also rest on the opinion itself (basis opinion_holding = hard, opinion_sentence = advice; opinion_says is the court's sentence), on an AI-written summary of the case (basis ai_holding / ai_headnote, advice only; ai_summary_says is machine-generated, not the court's words) or on citing courts' parentheticals, passages of the case that citing courts quote, or a machine-written summary of wh… [714 more characters not shown]
Annotations the server declares
- title: Audit every citation in a brief
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
check_citationVerify a legal citation actually exists before using it.Reads text others write1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Findings
- Info
The description is 8,183 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...h cited_for_match null = same subject by word overlap, direction not established."
Description, as the server wrote it Verify a legal citation actually exists before using it. CASES: returns the real case at that citation, its treatment flag, and whether the name matches: expected_case_name, or else a case name written in the citation itself ("Roe v. Wade, 410 U.S. 113"). This catches hallucinated or miscopied citations. A reporter cite that is not indexed comes back with a verdict, and the verdict is what to act on (MCP13_20260925): exists_unindexed = the case is REAL (a court in the corpus prints this exact citation, see cited_by_corpus, or the case is found by name and year, see name_candidates) and only our reporter index lacks it -- never call it a hallucination; not_indexed = no evidence either way, read volume_coverage: for 2019+ volumes of the state regional reporters we often index under 30% of the cases (coverage_pct), so a miss there is weak evidence; interior_mismatch / strong = the page provably sits inside a differently named case, or the volume is complete and nothing turns up by name -- treat as likely fabricated; name_mismatch = opinions in the corpus print this citation for a DIFFERENTLY named case (cited_by_corpus.examples[].printed_as): the cite is real but not the case you named -- fix the name or the cite. Local misses are also re-checked LIVE against CourtListener. STATUTES, REGULATIONS AND COURT RULES: a citation that is not a reporter cite is checked against the corpus that holds it (50 states + DC, 44 states' court rules, the Florida Administrative Code, USC/CFR/federal rules/USSG) and comes back with type, verified, the canonical citation, title, url and link_md; verified=false carries nearby sections and a coverage note, and is NEVER proof the section does not exist. ALWAYS use this to verify any citation that will appear in a legal document. With quote, the quoted words are checked against the opinion the cite resolves to (quote_check, the verify_quote block). PROPOSITION-LEVEL TREATMENT (MCP14_20260925): pass proposition (what you cite the case for; quote is used when proposition is absent) and treatment.proposition names the one proposition citing courts cite it for that matches, with its OWN flag, and treatment.proposition_note says when that point is flagged although the case is green (or unflagged although the case is red); "no proposition-level signal for this use" means only the case-level flag applies. PUBLIC-DOMAIN CITES AND ¶ PINPOINTS (MCP17_20260925): vendor-neutral cites are read as the courts print them ("2023 IL App (1st) 220520", "2023-Ohio-1234", "2022-NMCA-063", "2021 IL App (1st) 200563-U", "2019 S.D. 9"); a paragraph pinpoint ("¶ 45", "¶¶ 12-14", "at ¶ 7") on a resolved cite is checked against the opinion's own paragraph numbers: paragraph_pin.verdict interior_pin = that paragraph exists (paragraph_text shows it), interior_mismatch = past the last numbered paragraph, a number the opinion skips, or (with quote) the quotation sits in another paragraph (found_at, suggested); null = the text carries no paragraph numbers, not checked. The top-level verdict stays found. A verified statute, regulation or rule also returns currency + temporal_note (the edition held, current-through, last amendment where recorded); with as_of, temporal_warning comes first when the section changed after that date. OPERATIVE DATE: statute text is the edition this corpus holds, not the law on every date. For a crime, a contract, a limitations period or a procedural step, establish the operative date first (offense, contract, accrual or filing date) and pass it as as_of (YYYY-MM-DD); say so when the text postdates the facts, since an older offense may be charged under a predecessor section. A statute whose status is sunset, future, expired or repealed leads with status_warning (sunset_date, successor). So does a U.S. Code section omitted from the Code (status omitted). A citing opinion that says the cited case does not exist (a sanctions or show-cause order about fabricated citations, Mata v. Avianca) is never counted as evidence: such citers are listed in fabricated_citers, and when they are the only ones the verdict is strong with verdict_evidence cited_as_fabricated (likely fabricated). Clusters that look like one decision filed twice but share no docket number are listed as twin_candidates (not folded). A CFR section removed from the eCFR is status removed (status_warning REMOVED, currency.sunset_date = the removal's effective date). A case match with a non-green flag carries treatment.flag_reason and flag_basis. A FORMER U.S. Code appendix cite ("50 U.S.C. App. § 462", "46 U.S.C. App. § 688") comes back with status former_appendix (verified stays false): the appendix no longer exists, the cite is not fabricated, and former_appendix names the current section(s) from the Law Revision Counsel's disposition tables. A 5 U.S.C. App. 2 / 3 / 4 section (FACA, Inspector General Act, Ethics Act; moved into Title 5 in 2022) also carries former_appendix with its current section. With proposition= and a pinpoint in the citation, cited_for says what courts cite that page for and, when another page of the case fits the proposition clearly better, suggested_pin_cite (a suggestion to read, not a verdict). With quote= and a pinpoint page, pinpoint_check compares the pin with the page the quotation is on in the cited reporter: quote_on_other_page gives quote_page and a cite_as with that page. With proposition= (or an explanatory parenthetical), cited_for_check says when courts cite the case for the opposite, for a different amendment / statute / standard, or for nothing like it. A bare California Style Manual code name ("Code Civ. Proc. § 128.5", "Pen. Code § 496") is read as California by default; state="xx" reads it as another state's code; "art." forms stay Louisiana. Constitutions, "former" sections, bar rules, agency decisions and date + docket citations are read too; recognised_form true means the form is known and cannot be checked here, which is no signal it is wrong. For a California opinio… [2183 more characters not shown]
Annotations the server declares
- title: Check a citation
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: true
context_detectorDoes the cited case actually say what the brief says it says?No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Description, as the server wrote it Does the cited case actually say what the brief says it says? Call it after check_brief for every cite you rely on whose support_check says available, or for any sentence you want to test against the case it cites. Give the citation (reporter cite with its first page, or cluster_id), the sentence as written, and the pin page: it returns the case (bluebook, cluster_id, url), pinned_window = about 2,500 characters of the opinion around the pinned page's best word match, best_window = the same around the opinion's best match when that is another page, lexical overlap scores, and an instruction for judging the pair (supported / overstated / unsupported / cannot_tell). The windows are placed by word overlap; the judgement is yours. Case opinions only (statutes and rules are out of scope). Up to 12 cites in one call with items.
Annotations the server declares
- title: Lay the cited opinion next to the brief's sentence
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
export_documentOptional.No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Description, as the server wrote it Optional. If your client can save files locally and you want a readable copy of a case or statute, this returns the complete document to save; otherwise use get_case / get_statute as before. The document is the syfert.com page as the website renders it (caption, headings, footnotes, star-page markers such as [*551], links to the cited cases): Markdown with YAML front matter (default) or one self-contained HTML file. Returns filename, format, bytes, document and note; the MCP cannot write to disk, so saving document under filename is up to you. A case: cluster_id, or citation (+ case_name). A statute or rule: jurisdiction + citation, as get_statute takes them. A document longer than one response (about 20k tokens) comes in parts that each fit one (page / pages, truncated, next_offset): call again with offset = next_offset and append the parts in order.
Annotations the server declares
- title: Save a readable copy of a case or statute
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
suggest_treatmentCrowdsourced AI feedback on a Syfertize treatment flag.No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Description, as the server wrote it Crowdsourced AI feedback on a Syfertize treatment flag. When the evidence in front of you (an overruling, abrogating or reinstating opinion, or a later court holding the case is still good law) shows a case's flag is wrong, propose red (no longer good law) or green (still good law) with the SHORTEST possible one-line reason, ideally naming the authority. Only red or green is accepted, not yellow. Identify the case by cluster_id, or by citation (+ case_name). The reason must concern the case's legal status only, never the user's facts, question, client or document. Returns the current flag and what was recorded; a proposal equal to the live flag is not recorded. By calling this you help syfert.com improve its treatment flags through AI feedback. Nothing about your session, user, query or documents is logged: only the case id, the flag you propose, your one-line reason, the authority you name and the date are stored, for human review. Your suggestion does not change the live flag.
Annotations the server declares
- title: Suggest a red/green flag correction
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: false
verify_quoteCheck one quotation against the full text of the opinion it is attributed to, before it goes in quotation marks.No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Description, as the server wrote it Check one quotation against the full text of the opinion it is attributed to, before it goes in quotation marks. Returns verdict (verbatim | near | absent), percent, best_match (the opinion's own sentence, up to ~600 characters), star_page (the last *page marker before it, null when the text carries none), offset (get_case offset), case and advice. near means the opinion says it in other words: paste matched_span (the opinion's words over the span you quoted) instead; best_match is the sentence around it. absent means the words are not in this opinion. Identify the case by cluster_id or reporter citation (case_name helps a citation newer than the reporter index). Quotation marks mean pasted from a tool result (opinion text, passage, best_match). If you typed a quote from memory, run verify_quote and use best_match. Run check_brief on the finished draft and apply its corrections. Copy cite_as for the citation (first page, then pin). A case_name alone also works: find_case decides, and a name that is not decisive returns candidates instead of an answer.
Annotations the server declares
- title: Verify a quotation against the opinion
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
Only reads 14Show 14 read-only tools
find_authorityFind the case to cite for a proposition.1 finding
Why it is in this group
- The name "find_authority" says "find": a word for reading.
Findings
- Info
The description is 3,176 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...anonymous call gets the word-based rows only, with closeness measured on wording."
Description, as the server wrote it Find the case to cite for a proposition. Paste the proposition or the language you need authority for VERBATIM; do not convert it to keywords. The result names the source case to cite (source_case), the verbatim passage, and how many courts have adopted it (quoted_by_cases); match_kind says how it matched: exact_phrase (the passage contains your words), relaxed (it shares your distinctive words), proposition (a rule later courts cite that case for). A miss means no court has quoted or been cited for that wording; try find_issues for the doctrine, or search_cases with 3-4 distinctive words. Routing: a proposition or quoted language -> find_authority (pasted verbatim); a doctrine name or fact pattern -> find_issues; a party name, statute number or keywords -> search_cases. Quotation marks mean pasted from a tool result (opinion text, passage, best_match). If you typed a quote from memory, run verify_quote and use best_match. Run check_brief on the finished draft and apply its corrections. Copy cite_as for the citation (first page, then pin). The first 3 rows carry star_page and cite_as. A quotation inside the proposition (in quotation marks) is also searched on its own: passages that contain it word for word come first, match_kind quotation (matched_quotation names it). It also finds, by meaning, the PAGES later courts cite for a paraphrased proposition (match_kind cited_for_label: pin page, cite_as with the pin, a generated label of what the page is cited for, how many courts cite it, and the treatment of those citations), and returns string_cite: those pages one per case in Bluebook order with "(cited for …)" parentheticals to rewrite from the opinion. Up to 3 semantic_case rows at the end are whole opinions closest in meaning to the proposition: leads to read, not quotations. Every row carries match_sentence (the passage sentence or label that matched; left out when it is the whole quote or label) and closeness: verbatim | close | related, measured on wording and meaning (related = it shares the subject and may paraphrase the rule or only touch it: read it before citing); rows only loosely related are left out. no_close_authority: true means no row matches the proposition closely in wording: read the related rows' match_sentence before citing one, or use search_cases with distinctive words, find_issues, or a rephrased proposition. The default answer holds 15 rows (per_page up to 30): a case's third and later passages come after the other rows, and the first row of each kind and of the forum's highest court is always shown; rows_omitted counts the rest. compact=true returns short rows. A proposition that names a statute section ("section 83.49(3)", "42 U.S.C. § 1983") also gets up to 5 statute_construing rows: cases that cite that section, each with its own sentence citing it, ranked by closeness to the proposition (statutes_named lists the section and its get_statute arguments). Rows and grades found by meaning (cited_for_label, string_cite, semantic_case, closeness measured by meaning) are for signed-in callers (a free or Pro token); an anonymous call gets the word-based rows only, with closeness measured on wording.
Annotations the server declares
- title: Find the case to cite for a proposition
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
find_caseFast direct lookup: give a citation or a case caption and get THE case (or 'not decisive' with up to 5 candidates) without a full-text search.1 finding
Why it is in this group
- The name "find_case" says "find": a word for reading.
Findings
- Info
The description is 3,618 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...U.S. 436" (one name before the cite) is read as that cite with the name checked."
Description, as the server wrote it Fast direct lookup: give a citation or a case caption and get THE case (or 'not decisive' with up to 5 candidates) without a full-text search. Use search_cases for topics and words. A citation ("384 U.S. 436", "Miranda v. Arizona, 384 U.S. 436, 444 (1966)", "449 F. App'x 901", "2019-Ohio-10") resolves through the reporter index; when a case name is given (in the query or as expected_case_name) and the case at that citation carries a different name, the answer is NOT decisive (problem name_mismatch) and names the case that is really there. A caption ("Miranda v. Arizona", or one distinctive name such as "Celotex") is decisive when the top match is at least twice as strong as the runner-up, or when the case name IS the caption word for word (punctuation, company suffixes and abbreviations read) for one decision in the corpus (several filings of one decision count once: the answer is the Published, reported record, the others listed in also_filed_as) -- an exact caption outranks any opinion-text match; common captions ("Smith v. Jones", "State v. Johnson") are not decisive by design. A docket number ("2D09-5223", "No. 1:15-cv-00001" with court, "17-1618") resolves through the docket index: decisive when one case carries it, otherwise problem ambiguous_docket with the candidates (a Supreme Court number also matches its cert-stage entries). The answer carries case (bluebook, url, link_md, treatment_flag), kind (citation|caption|docket), decisive, candidates, and a note when not decisive. A red, yellow or procedural case also carries treatment_flag_reason and treatment_flag_basis. A Westlaw / Lexis number whose sources disagree about the case carries database_caution. For a caption that is not decisive, the candidates whose caption carries both parties (abbreviations and acronyms read) are listed first, marked matched_by caption, with caption_matches = how many decisions carry it. A misspelt or partial caption that nothing matches ("Palsgraph v. Long Island R.R.", "City of Keller") gets suggestions appended to candidates, marked matched_by caption_fuzzy (a close spelling of the party names, or a caption that begins with the one party given), with did_you_mean; they never make the answer decisive. A caption that is word for word ONE decision's is decisive even when other decisions share a party name. case carries disposition (disposition_source headnote) when a synthetic headnote exists. A California Court of Appeal opinion with a Supreme Court docket event carries citability {status depublished | review_granted | transferred | review_decided, date, s_number, source, rule Cal. R. Ct. 8.1115}; depublished / review granted / transferred lead with status_warning (a depublished opinion may not be cited; one under review has no binding effect, is citable for persuasive value and to establish a conflict in authority, and the cite must note the grant; one transferred with directions to vacate is generally not citable; order_says quotes the order when held); old_rule true = review granted before 2016-07-01, when the grant itself depublished the opinion (a later decision did not restore it): may_cite no, check the docket. Short forms (MCP101): "[Name, ]303 U.S. at 288" is read through the page run of the case that holds the page; for id., ibid. or supra pass sentence (the draft sentence containing it) and quotation: id. binds to the citation before it in that sentence, never to a case named earlier in the conversation. year and quotation break a tie among caption candidates. "Miranda, 384 U.S. 436" (one name before the cite) is read as that cite with the name checked.
Annotations the server declares
- title: Find one case by citation or caption
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
find_courtResolve a court name to court_id values usable in the search_cases court filter.No findings
Why it is in this group
- The name "find_court" says "find": a word for reading.
Description, as the server wrote it Resolve a court name to court_id values usable in the search_cases court filter. Also resolves whole GROUPS of courts (tribal courts, veterans, bankruptcy, all U.S. district courts) to the single aggregate token that filters on all of them: see court_groups in the result. Each court row carries bluebook, the Bluebook court abbreviation for a citation parenthetical (e.g. "Fla. 1st DCA"; "U.S." for the U.S. Supreme Court, whose case citations to U.S. / S. Ct. carry no court in the parenthetical; absent for courts with none on file); citation_string is the stored CourtListener string.
Annotations the server declares
- title: Find a court id
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
find_issuesENTRY BY ISSUE.Reads text others writeNo findings
Why it is in this group
- The name "find_issues" says "find": a word for reading.
What it can reach
- untrusted-inputThe name "find_issues" says "find" and "issue": it reads issues, which other people can write.
- untrusted-inputThe description's first sentence says "issue": it reads issues, which other people can write.
Description, as the server wrote it ENTRY BY ISSUE. Use this FIRST when the question arrives as facts or an issue rather than a case name ("can the appellate court affirm on a ground the trial court never reached", "impact rule emotional distress", "Fabre defendant"). Returns (1) the doctrine as courts in that STATE actually name it, with the cases on EACH side (followed_by / questioned_by, from how each citing opinion treated the case), example sentences judges wrote, the statutes those opinions construe, and every state where the phrase is alive with its year span (a doctrine retained in one state and abandoned elsewhere shows as a stalled span); and (2) the matching mined propositions (the rule text courts quote) with per-passage flags. Doctrine name is not case health: an issue can be alive while its anchor case is red. Read questioned_by before relying on followed_by. A q of more than 8 substantive words is read as a proposition: propositions (matched against all of it) come first. The default answer holds 3 issues and 5 statutes and stays under ~12,000 characters (issues_omitted counts the rest); max_issues or detail=true returns more. Routing: a proposition or quoted language -> find_authority (pasted verbatim); a doctrine name or fact pattern -> find_issues; a party name, statute number or keywords -> search_cases. Each issue also carries tree: where the doctrine sits in the Syfert issue tree (area > group > issue, built from what later courts cite pages for), how alive it is in the state, and related issues beside it. Browse the tree with node instead of q: find_issues(state="fl", node=0) lists the top areas cited by that state's courts; follow each child's next call down to an issue, which lists the pages courts cite for it.
Annotations the server declares
- title: Research an issue
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
get_brief_resultsSigned-in accounts only (a free account at https://syfert.com/mcp/; anonymous callers use check_brief with text).Reads text others writeNo findings
Why it is in this group
- The name "get_brief_results" says "get": a word for reading.
What it can reach
- untrusted-inputThe description's first sentence says "http": it reads web pages, which anyone can write.
Description, as the server wrote it Signed-in accounts only (a free account at https://syfert.com/mcp/; anonymous callers use check_brief with text). Read the results of check_brief(brief_id) one page at a time (each page under ~20k tokens). section: corrections (ranked by severity), cites (one full row per citation, as check_brief with detail="full"), quotes_unattached, statutes. Start without cursor, then pass next_cursor until it is null. After the last page of every section the session is deleted (session_deleted: true).
Annotations the server declares
- title: Page through a checked brief
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
get_caseFetch one case: metadata, parallel citations, treatment flag, and full opinion text (paginated via offset/max_chars).1 finding
Why it is in this group
- The name "get_case" says "get": a word for reading.
Findings
- Info
The description is 2,521 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...e is returned). With pin, the first passage is the pinned page's own best window."
Description, as the server wrote it Fetch one case: metadata, parallel citations, treatment flag, and full opinion text (paginated via offset/max_chars). cited_by is the number of opinions citing the case: CourtListener's count, or the syfert.com citation graph's when that finds more (cited_by_source: citation_graph says so); for a decision filed as several twin clusters it is summed over them (cited_by_note says so), and a duplicate or stub twin id is answered with its survivor (note says so). Citations inside the returned text are checked against the reporter index (embedded_citations); citation_warnings names any the court itself got wrong. Identify the case by cluster_id (from search_cases results) or by reporter citation. find_text returns passages: up to 5 sentence-bounded windows holding a phrase (or its closest match) with star_page and offset, so a long opinion need not be paged (pass include_text=false to get only those); quotable=true adds quotable_passages, the passages of this opinion other courts quote. Quotation marks mean pasted from a tool result (opinion text, passage, best_match). If you typed a quote from memory, run verify_quote and use best_match. Run check_brief on the finished draft and apply its corrections. Copy cite_as for the citation (first page, then pin). A California Court of Appeal opinion with a Supreme Court docket event carries citability {status depublished | review_granted | transferred | review_decided, date, s_number, source, rule Cal. R. Ct. 8.1115}; depublished / review granted / transferred lead with status_warning (a depublished opinion may not be cited; one under review has no binding effect, is citable for persuasive value and to establish a conflict in authority, and the cite must note the grant; one transferred with directions to vacate is generally not citable; order_says quotes the order when held); old_rule true = review granted before 2016-07-01, when the grant itself depublished the opinion (a later decision did not restore it): may_cite no, check the docket. focus = a question or proposition in plain words (no brief sentence needed): returns focus_passages, the 4 (focus_n, at most 8) passages of about 2,000 characters that bear on it most, each with page, offset, lexical score and part (majority / concurrence / dissent / syllabus ...), instead of reading from the opening; opinion text is then left out unless include_text=true. focus_miss=true means no passage matched (nothing else is returned). With pin, the first passage is the pinned page's own best window.
Annotations the server declares
- title: Read a case
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
get_citing_casesFull citator list: every ranked case citing this one, with judge-written parentheticals (usable verbatim in a brief), verbatim quotes, pin cites, and the citing court's own Blueboo…No findings
Why it is in this group
- The name "get_citing_cases" says "get": a word for reading.
Description, as the server wrote it Full citator list: every ranked case citing this one, with judge-written parentheticals (usable verbatim in a brief), verbatim quotes, pin cites, and the citing court's own Bluebook signal. Filter by flag color. Best for FINDING SUPPORTING AUTHORITY and parentheticals; for comprehensive negative-treatment checking use get_treatment (this list covers only each case's top-ranked ~50 citers).
Annotations the server declares
- title: Cases citing this case
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
get_propositionsWhat is this case actually cited FOR?No findings
Why it is in this group
- The name "get_propositions" says "get": a word for reading.
Description, as the server wrote it What is this case actually cited FOR? Returns the distinct legal propositions courts cite the case for (headnote-grade), each with its OWN treatment flag and counts. A case can be good law on one proposition and overruled on another. Use before citing a case for a specific point, and to find the strongest proposition to cite it for. passage is the opinion's own wording of the proposition with its star_page; passage null means citing courts paraphrase it. Pass proposition (your use of the case) to get proposition_match, the one proposition it matches, and proposition_note (MCP14_20260925). cited_for_pages (a separate array) lists pages later courts pin-cite, with a later court's parenthetical and, where it agrees with it, a machine-generated title that is never the court's words (PROPS229_20261110). Quotation marks mean pasted from a tool result (opinion text, passage, best_match). If you typed a quote from memory, run verify_quote and use best_match. Run check_brief on the finished draft and apply its corrections. Copy cite_as for the citation (first page, then pin).
Annotations the server declares
- title: What is this case cited for?
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
get_statuteFetch the full text of a statute, rule, or code section: the current codification in the edition this corpus holds.1 finding
Why it is in this group
- The name "get_statute" says "get": a word for reading.
Findings
- Info
The description is 2,132 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ch printed text of a Texas unit two acts amended without reference to each other."
Description, as the server wrote it Fetch the full text of a statute, rule, or code section: the current codification in the edition this corpus holds. Every answer carries currency (edition, current_through, current_through_basis, last_verified, text_last_changed = when our copy last changed, last_amended_year, effective_date, history_note, status, successor, versions_available) and temporal_note. OPERATIVE DATE: statute text is the edition this corpus holds, not the law on every date. For a crime, a contract, a limitations period or a procedural step, establish the operative date first (offense, contract, accrual or filing date) and pass it as as_of (YYYY-MM-DD); say so when the text postdates the facts, since an older offense may be charged under a predecessor section. With as_of, a section amended or enacted after that date returns temporal_warning FIRST, with the history note and, where this corpus kept it, the prior text (prior_text_available; most corpora keep none). A repealed or renumbered section returns status_warning with its successor. currency.status is current | sunset (operative until currency.sunset_date; successor names the version that takes over) | future (not yet operative; effective_date is when it starts) | expired | repealed | renumbered, and anything but current leads with status_warning. Where the later version is held (California sections with more than one version), as_of on or after its operative date returns that text in as_of_version. A U.S. Code section can also be omitted (dropped from the Code by the Office of the Law Revision Counsel and no longer printed; not a repeal): it leads with status_warning too, and the history note says why. A CFR section removed from the eCFR is status removed: status_warning REMOVED names the removal's effective date (currency.sunset_date) and the eCFR point-in-time link for the text in force before it. A former U.S. Code appendix section (usc, "50 U.S.C. App. § 462") returns status former_appendix with the current section(s) in former_appendix, not text. currency.competing_texts lists each printed text of a Texas unit two acts amended without reference to each other.
Annotations the server declares
- title: Read a statute or rule
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
get_treatmentIs this case still good law?1 finding
Why it is in this group
- The name "get_treatment" says "get": a word for reading.
Findings
- Info
The description is 2,325 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...h ambiguous_citation=true and candidates: pass case_name or cluster_id to choose."
Description, as the server wrote it Is this case still good law? Returns the Syfertize treatment flag (red = overruled/superseded, yellow = questioned/distinguished, green = followed, procedural = cert/rehearing denied) plus the citing cases behind each signal with quoted context. A case can be green overall and red on the proposition you need it for (MCP14_20260925): propositions_summary lists the propositions courts cite it for (most-cited first), each with its OWN flag; pass proposition (your use of the case) and proposition_match returns the one that matches, with its negative_citers, and treatment.proposition_note states the result ("no proposition-level signal for this use" = only the case-level flag applies). A red, yellow or procedural flag also carries treatment.flag_reason (why it is flagged, in one line, as the case page shows it), treatment.flag_basis (overruled_by | subsequent_history | scope | citing_events | curation_pin | twin_carry | higher_court_limit | validated_annotations) and treatment.flag_evidence. compact=true returns only the flag, scope, overruled_by and the top negative citers (about 90% smaller). When the annotation verb (overruled_by.verb) and the scope's verb differ, flag_reason leads with the scope's and carries the other in parentheses (overruled_by.verb_note). A case_name alone also works: find_case decides, and a name that is not decisive returns candidates instead of an answer. A California Court of Appeal opinion with a Supreme Court docket event carries citability {status depublished | review_granted | transferred | review_decided, date, s_number, source, rule Cal. R. Ct. 8.1115}; depublished / review granted / transferred lead with status_warning (a depublished opinion may not be cited; one under review has no binding effect, is citable for persuasive value and to establish a conflict in authority, and the cite must note the grant; one transferred with directions to vacate is generally not citable; order_says quotes the order when held); old_rule true = review granted before 2016-07-01, when the grant itself depublished the opinion (a later decision did not restore it): may_cite no, check the docket. A citation that two or more different cases carry (a table page, two decisions at one page) answers with ambiguous_citation=true and candidates: pass case_name or cluster_id to choose.
Annotations the server declares
- title: Is this case still good law?
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
search_casesFull-text search over 10.8M U.S.1 finding
Why it is in this group
- The name "search_cases" says "search": a word for reading.
Findings
- Info
The description is 2,358 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ed-in callers (a free or Pro token) only; an anonymous call gets the text search."
Description, as the server wrote it Full-text search over 10.8M U.S. court opinions (all states + federal, via syfert.com). Call this to find case law on a topic, locate a case by name, or find cases citing a statute. Routing: a proposition or quoted language -> find_authority (pasted verbatim); a doctrine name or fact pattern -> find_issues; a party name, statute number or keywords -> search_cases. Each hit carries bluebook and url (snippet up to 240 chars); a hit whose snippet quotes an earlier case carries passage_origin, the case to cite for that language. A q that is one citation or an "X v. Y" caption naming one case decisively also returns that case as direct_hit above the results (find_case does only that lookup). Query syntax: boolean AND/OR/NOT, "exact phrases", proximity (term1 w/5 term2), wildcards (neglig*), and field filters inside q (name:, judge:, syllabus:). Statute-style numbers like 83.49 are matched as citations to that statute. total may be null: a court or date filter makes the exact count too expensive to compute, and total_pages + total_note then carry the scale. partial: true means the time budget was hit; see partial_note. tier names the caller's plan; _source names the backend that served the search: primary = the dedicated index, secondary = the cloud index (the anonymous lane, or the fallback while the primary is unreachable). When no opinion holds every word of a long query, the answer may carry relaxed=true with relaxed_note and relaxed_terms_kept / relaxed_terms_dropped: its rows (closest_match) hold only the kept words. A q that is an "X v. Y" caption also lists, first on page 1, the decisions whose caption carries both parties (abbreviations and agency acronyms read: "Stone v. INS", "ANR Pipeline Co. v. FERC"), marked matched_by caption; the ranked rows follow unchanged. semantic: true ranks opinions by MEANING instead of words (a paraphrased fact pattern finds cases that share none of its words; rows carry similarity, no snippet; state, court, date and status=published filters apply); a text search with no hits is retried that way automatically (semantic_fallback: true). An answer that fell back (semantic_fallback: no opinion holds the words; relaxed: not every word) leads with notice. semantic: true and the semantic fallback run for signed-in callers (a free or Pro token) only; an anonymous call gets the text search.
Annotations the server declares
- title: Search case law
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
search_quotesFind every case that has quoted a phrase.No findings
Why it is in this group
- The name "search_quotes" says "search": a word for reading.
Description, as the server wrote it Find every case that has quoted a phrase. Phrase-matched search over 4.3M distinct passages quoted by two or more opinions, ranked by how many cases quote them. Use it to (a) find the true source of half-remembered language BEFORE attributing it, and (b) locate the canonical wording of a rule. Returns the passage, its source case, and adoption counts. When no passage has the exact phrase it falls back to passages sharing its distinctive words (match_kind relaxed), and it adds the mined propositions later courts cite a case for (match_kind proposition). Same tool as find_authority. Routing: a proposition or quoted language -> find_authority (pasted verbatim); a doctrine name or fact pattern -> find_issues; a party name, statute number or keywords -> search_cases. Quotation marks mean pasted from a tool result (opinion text, passage, best_match). If you typed a quote from memory, run verify_quote and use best_match. Run check_brief on the finished draft and apply its corrections. Copy cite_as for the citation (first page, then pin). The first 3 rows carry star_page and cite_as.
Annotations the server declares
- title: Trace a quoted passage
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
search_statutesSearch statutes, COURT RULES and codes by citation or topic, for EVERY U.S.No findings
Why it is in this group
- The name "search_statutes" says "search": a word for reading.
Description, as the server wrote it Search statutes, COURT RULES and codes by citation or topic, for EVERY U.S. state plus federal. For a specific state, pass state="wi" (any two-letter code, all 50 states + DC) and a topic or section number; add kind="rule" for that state's court rules or kind="statute" for its code. Without state it searches only the kinds named in kinds (default: Florida). Handles citation lookups ("83.49", "174.02") and topic queries ("dog bite strict liability"). Each result carries currency (text_last_changed and, where recorded, last_amended_year / effective_date / status / successor) and corpus_currency gives each corpus's edition and current-through date. OPERATIVE DATE: statute text is the edition this corpus holds, not the law on every date. For a crime, a contract, a limitations period or a procedural step, establish the operative date first (offense, contract, accrual or filing date) and pass it as as_of (YYYY-MM-DD); say so when the text postdates the facts, since an older offense may be charged under a predecessor section. (as_of is taken by get_statute and check_citation.)
Annotations the server declares
- title: Search statutes and rules
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
No private-data, untrusted-input, outbound path in this server alone: no tool is tagged private-data or outbound.
- Reads private data0
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write3
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out0
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 18:48 UTC, unchanged since Nov 6, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 8
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- https://mcp.syfert.com/.well-known/oauth-protected-resource/mcp
- Authorization servers
- https://mcp.syfert.com
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- yes
- Client ID metadata documents
- yes
- HighThe whole server
The endpoint served tools/list without credentials, and the list has 8 tools that change things (1 destructive, 7 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://mcp.syfert.com/mcp: tools/list returned 22 tools to an anonymous client, including brief_delete (destructive), check_citation (write), verify_quote (write), check_brief (write), suggest_treatment (write) and 3 more
Posture changes
- Nov 6, 2026
- Tools that change things, listed without credentials: 6 to 8
Findings
Matches of fixed checks, with the text they matched.
- Warnbrief_delete
The server marks "brief_delete" not destructive (destructiveHint: false), and its own name, description or input give the class "destructive".
What this check looks forMatched text
annotations: {"readOnlyHint":false,"destructiveHint":false,"idempotentHint":false,"openWorldHint":false}; The name "brief_delete" says "delete": a word for removing or ending something.
- Infosearch_cases
The description is 2,358 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ed-in callers (a free or Pro token) only; an anonymous call gets the text search."
- Infofind_case
The description is 3,618 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...U.S. 436" (one name before the cite) is read as that cite with the name checked."
- Infoget_case
The description is 2,521 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...e is returned). With pin, the first passage is the pinned page's own best window."
- Infocheck_citation
The description is 8,183 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...h cited_for_match null = same subject by word overlap, direction not established."
- Infoget_treatment
The description is 2,325 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...h ambiguous_citation=true and candidates: pass case_name or cluster_id to choose."
Show 4 more findingsShow fewer
- Infocheck_brief
The description is 6,714 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...are and stats.polarity_check_note says so; no finding on those means not checked."
- Infofind_authority
The description is 3,176 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...anonymous call gets the word-based rows only, with closeness measured on wording."
- Infoget_statute
The description is 2,132 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ch printed text of a Texas unit two acts amended without reference to each other."
- Infobrief_upload
The description mentions "credential".
What this check looks forMatched text
description: "...s are held in server RAM only, for your credential alone, and are deleted right after the..."
History
Changes to its tool definitions and to its remote endpoint, newest first.
Tools13 changed.~13 changedView diff
- Changed:check_briefChanges thingsdescriptioninputSchema
- Changed:check_citationChanges thingsdescriptioninputSchema
- Changed:find_authorityOnly readsdescriptioninputSchema
- Changed:find_caseOnly readsinputSchema
- Changed:find_issuesOnly readsinputSchema
- Changed:find_related_casesOnly readsinputSchema
- Changed:get_caseOnly readsdescriptioninputSchema
- Changed:get_citing_casesOnly readsinputSchema
- Changed:get_propositionsOnly readsdescriptioninputSchema
- Changed:get_treatmentOnly readsinputSchema
- Changed:search_casesOnly readsdescriptioninputSchema
- Changed:search_quotesOnly readsinputSchema
- Changed:verify_quoteChanges thingsinputSchema
Tools3 added, 11 changed.+3 added~11 changedView diff
- Added:brief_deleteDeletes or cancels
- Added:brief_uploadChanges things
- Added:get_brief_resultsOnly reads
- Changed:check_briefChanges thingsdescriptioninputSchema
- Changed:check_citationChanges thingsdescription
- Changed:export_documentChanges thingsdescription
- Changed:find_caseOnly readsdescription
- Changed:find_courtOnly readsdescription
- Changed:get_caseOnly readsdescription
- Changed:get_statuteOnly readsdescription
- Changed:get_treatmentOnly readsdescription
- Changed:search_casesOnly readsdescription
- Changed:search_statutesOnly readsinputSchema
- Changed:verify_quoteChanges thingsdescription
PostureRemote posture changed. Tools that change things, listed without credentials: 6 -> 8.View diff
- Tools that change things, listed without credentials: 6 to 8
Tools4 changed.~4 changedView diff
- Changed:check_citationChanges thingsdescriptioninputSchema
- Changed:find_caseOnly readsdescriptioninputSchema
- Changed:get_caseOnly readsdescriptioninputSchema
- Changed:get_treatmentOnly readsdescription
Install
Pinned to version 2.0.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http legal-research https://mcp.syfert.com/mcp --header 'Authorization: <your-authorization>'claude mcp add --transport http legal-research https://mcp.syfert.com/mcp --header 'Authorization: <your-authorization>'Run in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__legal-research__search_cases",
"mcp__legal-research__find_case",
"mcp__legal-research__get_case",
"mcp__legal-research__get_treatment",
"mcp__legal-research__get_propositions",
"mcp__legal-research__get_citing_cases",
"mcp__legal-research__find_related_cases",
"mcp__legal-research__find_court",
"mcp__legal-research__search_quotes",
"mcp__legal-research__find_authority",
"mcp__legal-research__find_issues",
"mcp__legal-research__search_statutes",
"mcp__legal-research__get_statute",
"mcp__legal-research__get_brief_results"
],
"ask": [
"mcp__legal-research__check_citation",
"mcp__legal-research__verify_quote",
"mcp__legal-research__check_brief",
"mcp__legal-research__suggest_treatment",
"mcp__legal-research__export_document",
"mcp__legal-research__context_detector",
"mcp__legal-research__brief_upload"
],
"deny": [
"mcp__legal-research__brief_delete"
]
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as legal-research. Review it before you use it.
Capability tags behind 3 rules
- mcp__legal-research__check_citation untrusted-input
- mcp__legal-research__find_issues untrusted-input
- mcp__legal-research__get_brief_results untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/com.syfert/legal-research/verify?hash=bcfea4a101080e44827f0593f74d8346c1fc3a05b7875d498222072c4c3ac295" | jq -e '.changed == false'Pinned hash bcfea4a10108, from Nov 20, 2026, 19:17 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/com.syfert/legal-research/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/com.syfert/legal-research)Source: MCP Registry. Synced Nov 6, 2026, 17:17 UTC.
Syfert Legal Research MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the Syfert Legal Research MCP server safe?
- No scan can promise that. What its definitions show: 22 tools, 8 of which can change, send or delete something, and 10 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the Syfert Legal Research MCP server delete data?
- Yes. 1 tool can delete, cancel or revoke something: brief_delete.
Can the Syfert Legal Research MCP server do things that cannot be undone?
- No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.
What tools does the Syfert Legal Research MCP server have?
- 22 tools: 1 destructive, 7 write, 14 read. brief_delete, check_citation, verify_quote, check_brief, suggest_treatment, export_document, context_detector, brief_upload and 14 more.
Does the Syfert Legal Research MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the Syfert Legal Research MCP server send data to outside services?
- None of its tools sends, posts or publishes, or takes a recipient, channel or URL.
Has the Syfert Legal Research MCP server changed its tools?
- Yes, 3 times since Anlyon Radar started tracking it, most recently on Nov 20, 2026. Every change is on this page and in its Atom feed.