protein-mcp-server
io.github.cyanheads/protein-mcp-servercyanheadsstreamable-httpv0.9.1Scanned
Federated protein structure & annotation across experimental (PDB) and predicted (AlphaFold) models.
7 tools: 1 change things and 6 only read.
- Reads private data · 2
- Reads text others write · 7
- Can send data out · 0
- 2 findings
Tools listed without sign-in. Remote posture
With Anlyon
1 tool here changes 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 protein-mcp-server can do
The protein-mcp-server is published by the GitHub account cyanheads. It has 7 tools. It can create or change data (1 tool: protein_track_ligands), but cannot delete data. 6 tools only read.
2 tools read private data (protein_get_structure and protein_find_similar); 7 tools bring in text other people can write (protein_search_structures, protein_get_structure, protein_find_similar and 4 more).
The fixed checks matched 2 findings (long-description). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in.
Last scanned Nov 14, 2026, version 0.9.1.
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.
7 tools
Changes things 1Claude Code: ask
protein_track_ligandsLigand discovery and binding-site analysis across the PDB.Reads text others writeNo 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.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Ligand discovery and binding-site analysis across the PDB. mode "find_ligand" resolves a name or formula to chemical component IDs with metadata (formula, weight, SMILES), ranked by deposition frequency — most-deposited component first, so the top hit is the most common match for the name, not necessarily an exact name-string match. The ranking runs in blocks of max(limit, 25) candidates in RCSB's name-match order: start pages through the ranked blocks (keep limit fixed while paging), and totalCount and candidatesConsidered report how many components matched and how many the page ranked. mode "structures_with_ligand" returns PDB entries holding a bound copy of a ligand (by exact component ID — get the ID from find_ligand first), highest-resolution first, each with its resolution in Å, title, method, and the polymer entities whose chains carry the ligand. This mode and find_ligand's depositionCount count bound copies only: a component present only as a residue inside a polymer chain (e.g. MSE in a selenomethionine protein) or a sugar inside a branched glycan is not counted. mode "binding_site" returns the polymer residues lining a ligand's pocket in a given structure — amino acids, or nucleotides where the ligand binds DNA or RNA — with contact distances, each numbered in both the mmCIF label namespace (asymId, seqId) and the author namespace (authAsymId, authSeqId) used by deposited coordinates and most literature. Binding sites are experimental-only (computed from deposited coordinates; predicted models carry no bound ligands) and cover non-polymer ligands only: a peptide-like ligand deposited as a polymer chain (a BIRD molecule such as cyclosporin A in 1CWA) has none, and a miss names the components and polymer-chain ligands the entry does hold.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
Only reads 6Show 6 read-only tools
protein_analyze_collectionProfile the PDB into distributions and trends over an optional scoping query: counts by method, organism, or polymer composition; resolution and molecular-weight histograms; releas…Reads text others writeNo findings
Why it is in this group
- The name "protein_analyze_collection" says "analyze": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Profile the PDB into distributions and trends over an optional scoping query: counts by method, organism, or polymer composition; resolution and molecular-weight histograms; release-year timelines; and multidimensional cross-tabs (e.g. method × release_year). Aggregation runs at RCSB, so the response carries compact counts per bucket rather than the matching entries. Pass one group_by dimension for a single breakdown, or two distinct dimensions for a cross-tab (the first nests the second). bucket_limit caps each dimension level separately rather than the response, so a cross-tab returns up to that many nested buckets under each of its capped parent buckets; bucketsReturned reports the realized total.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
protein_compare_structuresStructurally align multiple structures (up to the configured batch cap; structures past it are returned in ignored) via the RCSB Structural Comparison service (TM-align / jFATCAT).Reads text others writeNo findings
Why it is in this group
- The name "protein_compare_structures" says "compare": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Structurally align multiple structures (up to the configured batch cap; structures past it are returned in ignored) via the RCSB Structural Comparison service (TM-align / jFATCAT). reference:"first" aligns every structure to the first; reference:"all_pairs" computes the full pairwise matrix. Each pair is an independent async alignment job with per-pair partial success — a pair still computing when its poll budget or the call's 50 s budget runs out returns status "computing" with its job UUID, and a failed pair degrades its row without sinking the others, keeping its job UUID when the job was submitted. Re-call with a matching entry in resume[] to poll a pair's job UUID instead of resubmitting; a resumed pair reports a and b in the order its job was submitted, and a resume under a different method is rejected. Returns TM-score, RMSD, and aligned-residue count per pair, plus each structure's modeled-residue count and alignment coverage. TM-score is length-normalized and can shift sharply between structures that differ only by a terminal residue or two — the greedy superposition can settle into a worse local optimum — so read tmScore alongside rmsd, alignedResidues, modeledResidues and coverage, the columns that make such cases diagnosable.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
protein_find_similarFind structurally or evolutionarily related proteins.Reads private dataReads text others write1 finding
Why it is in this group
- The name "protein_find_similar" says "find": a word for reading.
What it can reach
- private-dataThe input names a database (input.databases), and the tool reads: it reads databases.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Findings
- Info
The description is 2,045 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ame maps to a PDB entry or UniProt accession that protein_get_structure resolves."
Description, as the server wrote it Find structurally or evolutionarily related proteins. by:"sequence" runs an RCSB mmseqs2 sequence-similarity search (synchronous; experimental PDB entries only — protein_search_structures with a sequence also reaches computed models) over a sequence — supplied directly, taken from one protein entity of a PDB entry (chain picks it; by default the lowest-numbered protein entity of at least 10 residues; queryEntityId names it), or pulled from a UniProt accession (isoforms included) or entry name (HBA_HUMAN); each hit carries its identity, E-value, bit score, and aligned range. by:"structure" runs a Foldseek fold-similarity search (asynchronous) against experimental and predicted databases, uploading the PDB entry's mmCIF file or the accession's AlphaFold DB model, up to 16,000,000 characters; if the job is still computing when the poll budget elapses, or the call's 50 s budget runs out after Foldseek accepted the job, the response reports status "computing" with a ticket — re-call with ticket_id set to that value to resume the same job instead of resubmitting, and a complete response carries the same ticket so a finished job can be re-paged with a different start. Foldseek searches each chain of a multichain structure as its own query: a response covers one query (query, default 0), reports queryCount, and ranks that query's hits by score across all searched databases. Each mode reads only its own controls: sequence, chain, max_evalue and min_identity belong to by:"sequence"; ticket_id, databases and query belong to by:"structure"; pdb_id, uniprot, start and limit are shared. A field the selected mode cannot consume is rejected rather than silently ignored. A UniProt accession merged into another entry answers for its successor with a notice; a demerged or deleted one fails with its replacements. Output names the engine and database each hit came from; a structure hit also carries its Foldseek target name, and an id only where that name maps to a PDB entry or UniProt accession that protein_get_structure resolves.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
protein_get_annotationsSequence and functional annotation for a protein: UniProt positional features (domains, binding and active sites, PTMs including cross-links and lipidation, topology, DNA binding, …Reads text others writeNo findings
Why it is in this group
- The name "protein_get_annotations" says "get": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Sequence and functional annotation for a protein: UniProt positional features (domains, binding and active sites, PTMs including cross-links and lipidation, topology, DNA binding, processing products such as chains and signal peptides, mutagenesis), natural variants, and InterPro domain/family memberships (Pfam, PROSITE, …) with GO terms. Provide a canonical UniProt accession directly, or a PDB ID — it is resolved to its UniProt accession via the structure's sequence cross-reference. A multi-chain PDB entry can map to several accessions; the default pick is deterministic (lowest author chain ID) and the alternatives are listed in "ambiguity" — pass "chain" to select a specific one. An accession UniProt merged into another entry answers for that successor with a notice; a demerged or deleted one fails with its replacement accessions. Use "include" to scope which annotation classes are fetched. Every response carries an "attribution" block with the upstream data licenses and citations.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
protein_get_structureFetch structures with metadata and coordinate-file URLs.Reads private dataReads text others write1 finding
Why it is in this group
- The name "protein_get_structure" says "get": a word for reading.
What it can reach
- private-dataThe description's first sentence says "fetch" and "file": it reads files.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Findings
- Info
The description is 2,456 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...tructure to page through its file. coordinateUrls always link the complete files."
Description, as the server wrote it Fetch structures with metadata and coordinate-file URLs. source "experimental" takes PDB entry IDs, and also resolves the computed-model IDs protein_search_structures returns (AF_*/MA_*), which come back marked source "predicted" with their modelling provider; "predicted" takes UniProt accessions, isoform accessions (P04637-2), or AlphaFold DB entry IDs (AF-P04637-2-F1) (AlphaFold, with pLDDT/PAE confidence); "best_available" takes the same identifiers, an entry ID reduced to its accession, and returns the top federated model — the PDB entry with the widest UniProt coverage, ties going to the finest resolution (coverage measures the mapped span, so the pick can be an engineered mutant over the wild-type entry), else the model with the highest confidence on the lDDT scale (pLDDT/100 or QMEANDisCo), unscored models last, which can include an experimental model outside the PDB such as a SASBDB fit; best_available records carry coverage and the UniProt span. Under predicted and best_available, a UniProt accession merged into another entry resolves to its successor's model with a notice naming the merge, and a demerged or deleted accession lands in failed[] with its replacements. Records fetched under source "experimental", computed models included, also carry polymer entities with both chain namespaces (labelAsymIds for protein_compare_structures, authAsymIds for protein_get_annotations; each lists an entity's first 100 chains beside its chainCount, or every chain with all_chains), bound ligands, molecular weight, and release date. Resolves up to the configured batch cap per call with per-ID partial success: IDs are trimmed and repeats looked up once, missed IDs are listed in failed[] with their reasons, and IDs beyond the cap are listed in ignored, never looked up. Upstream lookups share one time budget per call; under predicted and best_available, an ID not finished within it, or whose upstream lookup failed, lands in failed[] with a reason to re-call it. Set include_coords to inline coordinate text, held to 24000 characters per response on both surfaces: a lone selected file arrives in 24000-character windows, each giving coordinatesNextOffset to pass back as coords_offset; several selected files inline whole when they total 24000 characters or fewer, and otherwise overflow lists each file's length — re-call with sections naming one structure to page through its file. coordinateUrls always link the complete files.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
protein_search_structuresSearch experimental (PDB) and predicted (computed-model) protein structures by free text, protein sequence (triggers an mmseqs2 similarity search), and/or organism, method, resolut…Reads text others writeNo findings
Why it is in this group
- The name "protein_search_structures" says "search": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Search experimental (PDB) and predicted (computed-model) protein structures by free text, protein sequence (triggers an mmseqs2 similarity search), and/or organism, method, resolution, and bound-ligand filters. Returns ranked hits; the experimental page is enriched with title, method, resolution, and source organisms (for a sequence hit, those of the matched entity), and every sequence hit carries its identity, E-value, bit score, and aligned range against the query. Chain hit IDs into protein_get_structure. Optionally returns a facet breakdown (counts by method / organism / release year / …) alongside the hits in the same response. Facets count the filtered result set, so an active organism or method filter also constrains its own facet: the facet lists only the values found in the matching entries (other organisms in the same complexes, other methods deposited with the same entries), never the alternatives the filter excluded. Numeric histogram buckets (resolution, molecular weight) carry explicit rangeFrom/rangeTo bounds so a boundary label is unambiguous.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
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 outbound.
- Reads private data2
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write7
- 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 15, 2026, 13:19 UTC, unchanged since Nov 15, 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
- 1
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
- WarnThe whole server
The endpoint served tools/list without credentials, and the list has 1 tool that changes things (1 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://protein.caseyjhand.com/mcp: tools/list returned 7 tools to an anonymous client, including protein_track_ligands (write)
Findings
Matches of fixed checks, with the text they matched.
The description is 2,456 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...tructure to page through its file. coordinateUrls always link the complete files."
The description is 2,045 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ame maps to a PDB entry or UniProt accession that protein_get_structure resolves."
History
Changes to its tool definitions and to its remote endpoint, newest first.
No change seen yet. Its tools are rescanned every day, and a remote endpoint is looked at once a week.
Install
Pinned to version 0.9.1. Values in angle brackets are yours to fill in.
claude mcp add --transport http protein https://protein.caseyjhand.com/mcpclaude mcp add --transport http protein https://protein.caseyjhand.com/mcpRun 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__protein__protein_search_structures",
"mcp__protein__protein_get_structure",
"mcp__protein__protein_find_similar",
"mcp__protein__protein_compare_structures",
"mcp__protein__protein_analyze_collection",
"mcp__protein__protein_get_annotations"
],
"ask": [
"mcp__protein__protein_track_ligands"
],
"deny": []
}
}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 protein. Review it before you use it.
Capability tags behind 7 rules
- mcp__protein__protein_search_structures untrusted-input
- mcp__protein__protein_get_structure private-data, untrusted-input
- mcp__protein__protein_find_similar private-data, untrusted-input
- mcp__protein__protein_track_ligands untrusted-input
- mcp__protein__protein_compare_structures untrusted-input
- mcp__protein__protein_analyze_collection untrusted-input
- mcp__protein__protein_get_annotations 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/io.github.cyanheads/protein-mcp-server/verify?hash=22eae27b8755e9b5fbef174b932585750418cc61d2ccccc0c289cc663699df46" | jq -e '.changed == false'Pinned hash 22eae27b8755, from Nov 14, 2026, 22:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/io.github.cyanheads/protein-mcp-server/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/io.github.cyanheads/protein-mcp-server)Source: MCP Registry · repository. Synced Nov 14, 2026, 23:17 UTC.
protein-mcp-server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the protein-mcp-server safe?
- No scan can promise that. What its definitions show: 7 tools, 1 of which can change, send or delete something, and 2 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 protein-mcp-server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the protein-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 protein-mcp-server have?
- 7 tools: 1 write, 6 read. protein_track_ligands, protein_search_structures, protein_get_structure, protein_find_similar, protein_compare_structures, protein_analyze_collection and protein_get_annotations.
Does the protein-mcp-server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the protein-mcp-server send data to outside services?
- None of its tools sends, posts or publishes, or takes a recipient, channel or URL.
Has the protein-mcp-server changed its tools?
- Not since Anlyon Radar started tracking it. Last scanned Nov 14, 2026.