Sofya MCP server
co.sofya/sofyasofya.costreamable-httpv1.32.0Scanned
Web search, fetch, extract, and research for AI agents. Markdown output + AI-synthesized answers.
4 tools: 1 change things and 3 only read.
- Reads private data · 0
- Reads text others write · 3
- Can send data out · 2
- No findings
Sign-in required, OAuth metadata published. 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 Sofya MCP server can do
The Sofya MCP server is published by sofya.co, a domain the MCP Registry verified. It has 4 tools. It can create or change data (1 tool: research), but cannot delete data. 3 tools only read.
3 tools bring in text other people can write (search, fetch and extract); 2 tools can send data out of your system (fetch and extract).
None of the fixed checks matched anything in its tool definitions. That is not a safety guarantee.
Its remote endpoint asks for sign-in and publishes OAuth metadata.
Last scanned Oct 21, 2026, version 1.32.0.
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.
4 tools
Changes things 1Claude Code: ask
researchPerform comprehensive research on a topic.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 Perform comprehensive research on a topic. Decomposes your query into sub-queries, searches and reads multiple sources in parallel, then synthesizes a structured report with citations. Best for open-ended or comparative questions that need coverage from many angles. For simple factual lookups, use search instead (optionally with include_answer=true for cheap synthesis). Costs 50 credits. Returns: query, report (structured markdown with citations), sources (array of {title, url, fetched}), sub_queries (the decomposed queries), credits_used, credits_remaining, usage (token counts). Args: query: The research question or topic topic: "general" (default) or "news" (prioritize recent news articles) freshness: Filter by recency - "day", "week", "month", "year", or "YYYY-MM-DD:YYYY-MM-DD" max_sources: Maximum number of sources to use, 5-30 (default 20)
Only reads 3Show 3 read-only tools
extractFetch a webpage and extract specific information using AI.Reads text others writeCan send data outNo findings
Why it is in this group
- The description's first verb is "fetch": a word for reading.
What it can reach
- untrusted-inputThe description's first sentence says "fetch" and "webpage": it reads web pages, which anyone can write.
- untrusted-inputThe input takes a URL (input.url), and the tool reads it: it reads web pages, which anyone can write.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Claude Code: ask, because it is tagged outbound.
Description, as the server wrote it Fetch a webpage and extract specific information using AI. Use this when you need one specific answer from a page (e.g. a price, a spec, contact info) rather than its full content. The answer comes back as text, not JSON. Costs 10 credits. If the page has no usable text (empty or JavaScript-rendered body), the model is NOT called: content comes back empty and usage.low_content is true, rather than a fabricated answer. Gate on usage.low_content (or usage.content_chars) to detect pages you cannot ground on. Returns: content (the extracted text), url, credits_used, credits_remaining, usage (input_tokens, output_tokens, content_chars, low_content). Args: url: The URL to extract from prompt: What information to extract (e.g. "list all pricing tiers with features" or "extract the author name and publication date")
fetchFetch one or more URLs and return their content as clean markdown.Reads text others writeCan send data outNo findings
Why it is in this group
- The name "fetch" says "fetch": a word for reading.
What it can reach
- untrusted-inputThe input takes a URL (input.urls), and the tool reads it: it reads web pages, which anyone can write.
- outboundThe input takes a URL (input.urls): a request to a URL can carry data to whoever runs it.
Claude Code: ask, because it is tagged outbound.
Description, as the server wrote it Fetch one or more URLs and return their content as clean markdown. Use this to read articles, documentation, blog posts, or any page where you need the complete text, not just a snippet from search. Also supports PDF, DOCX, and other document formats. Costs 2 credits per URL. Max 10 URLs per request. Failed URLs are not charged. Set include_raw_html=true to also get the raw HTML source in each result. Useful for inspecting embedded URLs, data attributes, iframes, or script tags that are stripped during markdown conversion. May be null for non-HTML content (PDF, DOCX, etc.) and for some sites. Same cost. Returns: results (array of {title, url, content, raw_html, published_time, success, error}), credits_used, credits_remaining. Args: urls: List of URLs to fetch (max 10) include_raw_html: Include raw HTML source in each result (default false)
searchSearch the web for current information on any topic.Reads text others writeNo findings
Why it is in this group
- The name "search" says "search": a word for reading.
What it can reach
- untrusted-inputThe description's first sentence says "search" and "web": it reads web pages, which anyone can write.
Description, as the server wrote it Search the web for current information on any topic. Returns extracted page content, not just snippets. Best for factual lookups, specific questions, or when you need a list of sources. For open-ended questions that need synthesis across many sources, use the research tool instead. For news queries (current events, breaking news, politics, world events), set topic="news" to search news sources specifically. This returns recent articles with publication dates. Set include_answer=true to get an AI-synthesized answer alongside results (adds 10 credits). This is the sweet spot for most agent tasks, e.g. basic + include_answer = 12 credits, much cheaper than a full 50-credit research call. Returns: query (the query actually run), answer (if requested), results (array of {title, url, content, description, fetched, published_date, sublinks, table}), search_depth, topic, elapsed_ms, credits_used, credits_remaining, altered_query, relaxed_query (set when the query matched nothing and was retried once with its site: operator, else its quotes, removed - the results answer that looser query). Args: query: The search query search_depth: "basic" (default) for extracted page content (2 credits), "snippets" for SERP snippets only without page fetching (1 credit) max_results: Number of results (default 10, max 20) include_answer: Generate an AI answer that synthesizes the search results (adds 10 credits) include_domains: Only include results from these domains, as bare hostnames like "github.com" (max 10) exclude_domains: Exclude results from these domains, as bare hostnames (max 10) topic: "general" for web search, "news" for news articles. use "news" for current events, breaking news, politics, or any time-sensitive query freshness: Filter by recency - "day", "week", "month", "year", or "YYYY-MM-DD:YYYY-MM-DD"
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.
- 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.
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, 18:32 UTC, unchanged since Oct 29, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 401
- tools/list served without credentials
- no
- Tools that change things, listed without credentials
- 0
- Protocol version
- not observed
- WWW-Authenticate on 401
- yes
- resource_metadata in WWW-Authenticate
- https://mcp.sofya.co/.well-known/oauth-protected-resource
- RFC 9728 protected resource metadata
- https://mcp.sofya.co/.well-known/oauth-protected-resource
- Authorization servers
- https://mcp.sofya.co
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- yes
- Client ID metadata documents
- no
Posture changes
- Oct 29, 2026
- HTTP status of an anonymous initialize: 200 to 401
- tools/list served without credentials: yes to no
- Tools that change things, listed without credentials: 1 to 0
- Protocol version: 2025-03-26 to none
- WWW-Authenticate on 401: not observed to yes
- resource_metadata in WWW-Authenticate: none to https://mcp.sofya.co/.well-known/oauth-protected-resource
- RFC 9728 protected resource metadata: https://sofya.co/.well-known/oauth-protected-resource/mcp to https://mcp.sofya.co/.well-known/oauth-protected-resource
- Authorization servers: https://sofya.co to https://mcp.sofya.co
Findings
Matches of fixed checks, with the text they matched.
No check matched anything in these tool definitions. That is not a safety guarantee.
History
Changes to its tool definitions and to its remote endpoint, newest first.
PostureRemote posture changed. HTTP status of an anonymous initialize: 200 -> 401; tools/list served without credentials: yes -> no; Tools that change things, listed without credentials: 1 -> 0 (and 5 more).View diff
- HTTP status of an anonymous initialize: 200 to 401
- tools/list served without credentials: yes to no
- Tools that change things, listed without credentials: 1 to 0
- Protocol version: 2025-03-26 to none
- WWW-Authenticate on 401: not observed to yes
- resource_metadata in WWW-Authenticate: none to https://mcp.sofya.co/.well-known/oauth-protected-resource
- RFC 9728 protected resource metadata: https://sofya.co/.well-known/oauth-protected-resource/mcp to https://mcp.sofya.co/.well-known/oauth-protected-resource
- Authorization servers: https://sofya.co to https://mcp.sofya.co
Install
Pinned to version 1.32.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http sofya https://mcp.sofya.co/mcpclaude mcp add --transport http sofya https://mcp.sofya.co/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__sofya__search"
],
"ask": [
"mcp__sofya__fetch",
"mcp__sofya__extract",
"mcp__sofya__research"
],
"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 sofya. Review it before you use it.
Capability tags behind 3 rules
- mcp__sofya__search untrusted-input
- mcp__sofya__fetch untrusted-input, outbound
- mcp__sofya__extract untrusted-input, outbound
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/co.sofya/sofya/verify?hash=bc2f9aec4da7ab45084464e83e2342d43dfc439926c13abdce2e8bb3c95eb99b" | jq -e '.changed == false'Pinned hash bc2f9aec4da7, from Oct 21, 2026, 18:32 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/co.sofya/sofya/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/co.sofya/sofya)Source: MCP Registry · repository. Synced Nov 15, 2026, 17:17 UTC.
Sofya MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the Sofya MCP server safe?
- No scan can promise that. What its definitions show: 4 tools, 1 of which can change, send or delete something, and 0 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 Sofya MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the Sofya 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 Sofya MCP server have?
- 4 tools: 1 write, 3 read. research, search, fetch and extract.
Does the Sofya MCP server need sign-in or OAuth?
- Yes. Its remote endpoint answered 401 without credentials, and it publishes OAuth protected resource metadata (RFC 9728); its authorization server supports dynamic client registration.
Does the Sofya MCP server send data to outside services?
- 2 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): fetch and extract.
Has the Sofya MCP server changed its tools?
- Not since Anlyon Radar started tracking it. Last scanned Oct 21, 2026.