Skip to main content
Glama
anonym-community-mcp

anonym-community-mcp

by georgecurta

anonym-community-mcp

Read-only Model Context Protocol server for the anonym.community privacy research corpus — 1,478 documented PII/privacy pain points, 98 structural root causes, a 240-jurisdiction privacy-law directory, 134 evidence-backed FAQ answers, and 1,600+ cited academic papers, queryable directly from Claude Desktop, Claude Code, Cursor, or any MCP client.

License: MIT Node.js >= 18 Self-test MCP

Why this exists

All of this is already public as HTML and JSON at anonym.community, but an agent trying to answer "which EU jurisdictions have a data protection authority and what law applies" has to crawl and re-derive that from prose every time. This server exposes the same corpus as typed, queryable tools, so an agent can ask precisely and cite exactly — pain point IDs, jurisdiction records, FAQ entries and paper DOIs, not paraphrased HTML.

Related MCP server: WisdomAgent

What it exposes

Tool

Data

Use it for

search_pain_points

1,478 documented PII/privacy problems

"what goes wrong with X"

list_structural_drivers

98 irreducible root causes

"why does this keep happening"

lookup_jurisdiction

240 jurisdictions, DPAs, legislation

"is there a privacy law in X"

search_faq

134 evidence-backed Q&As

direct practitioner questions

search_research_papers

1,600+ academic papers with DOIs

"cite a primary source"

get_corpus_stats

counts, valid filter values, data freshness

call first to discover valid track/region values

Every result carries its identifiers and source links, so an answer can be cited back to a specific pain point, driver, jurisdiction, FAQ entry or DOI.

Install

Claude Desktop / Claude Code

Add to your MCP config (claude_desktop_config.json, or .mcp.json in a project):

{
  "mcpServers": {
    "anonym-community": {
      "command": "npx",
      "args": ["-y", "github:georgecurta/anonym-community-mcp"]
    }
  }
}

No install step, no clone, no local path. npx fetches and runs it on demand.

If you'd rather run a pinned local copy:

git clone https://github.com/georgecurta/anonym-community-mcp.git
{
  "mcpServers": {
    "anonym-community": {
      "command": "node",
      "args": ["/absolute/path/to/anonym-community-mcp/index.js"]
    }
  }
}

Cursor / other MCP clients

Same shape — the server speaks stdio JSON-RPC, the MCP default. No port, no credentials, nothing to host.

Verify it works

node index.js --selftest

Runs 7 assertions against the live corpus (not fixtures) and exits non-zero on failure. Expected output ends with 7 passed, 0 failed.

Manual protocol check:

printf '%s\n%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"probe","version":"1"}}}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list"}' | node index.js

Example

{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
  "name":"search_pain_points",
  "arguments":{"query":"data broker opt-out","limit":2}
}}

returns matched pain points with title, evidence, severity, structural driver track, category and source citations — ready to quote, not to re-summarize.

Architecture

Live, not bundled. The server fetches https://anonym.community/data/*.json — the same files the website itself is built from — at startup, caches in memory, and refreshes on a TTL (stale-while-revalidate: a request just after the TTL expires gets the cached answer immediately while a background fetch updates the cache for the next one). The corpus this server answers from is never more than ANONYM_MCP_TTL_MS (default 30 minutes) old, with no sync step to remember and no risk of a stale local snapshot drifting from the live site.

Env var

Default

Purpose

ANONYM_MCP_BASE_URL

https://anonym.community

Point at a staging mirror instead

ANONYM_MCP_TTL_MS

1800000 (30 min)

How long a cached fetch is served before refreshing

Read-only by construction. There are no write tools, no filesystem writes and no network calls other than the GET requests above. Data is fetched, held in memory, and never mutated — locally or remotely. The server has no code path that could alter the site, so it cannot be talked into doing so.

stdio, not HTTP. A hosted HTTP/SSE endpoint would make the corpus reachable by remote agents, but it also means another public service to patch and rate-limit. stdio gives the same query capability with no attack surface and no operational burden. handle(name, args) is exported and transport-agnostic if a hosted endpoint is ever wanted — wrap it, don't rewrite it.

No SDK. The MCP subset actually needed here — initialize, tools/list, tools/call — is implemented directly against the JSON-RPC 2.0 spec, so the repo has zero runtime dependencies beyond Node's own fetch.

Data license

The code in this repository is MIT-licensed (see LICENSE). The corpus it queries — anonym.community's pain points, structural drivers, jurisdictions, FAQ and paper citations — is published separately under CC-BY-4.0 by curta.solutions. Attribute the source when citing results.

anonym.legal

The product most directly related to this repo. anonym.legal is a cloud/desktop PII anonymization platform (267+ entity types, 48 languages, Zero-Knowledge architecture) built on the same research this corpus documents — and it ships its own MCP server (7 tools for Claude Desktop and Cursor) for actually anonymizing data, as distinct from this repo, which only lets an agent read the research about why anonymization matters. If a query against this corpus surfaces a pain point your own data has, anonym.legal is the tool built to act on it.

Issues

Found a data problem, a broken query, or a gap in what's exposed? Open an issue — corrections to the underlying corpus itself go through anonym.community/contact.html instead, since this repo only serves what that site publishes.

License

MIT © George Curta / curta.solutions

Available Tools

6 tools
get_corpus_statsA

Return counts, available filter values (tracks, regions), and data freshness for this corpus. Call this first to discover valid track and region values for the other tools.

Inputs: no named parameters.

list_structural_driversA

List or search the 98 structural drivers — the irreducible root causes that generate the pain points. Breaking one driver weakens many pain points at once, so these are the right unit for "why does this keep happening" and "what should we actually fix" questions.

Inputs: limit, query, track.

lookup_jurisdictionA

Look up privacy-law status for a country or region across 240 jurisdictions: whether it has a data protection authority, what legislation applies, and links. Use for "is there a privacy law in X" and cross-border transfer questions.

Inputs: limit, region, country.

search_faqA

Search 134 evidence-backed privacy questions and answers, each with root cause, real-world example and supporting data points. Use for direct practitioner questions.

Inputs: query (required), limit.

search_pain_pointsA

Search 1,478 documented PII/privacy pain points. Each is a concrete, evidenced problem (e.g. re-identification via quasi-identifiers, opt-out futility with data brokers), tagged with research track, category and severity. Use for "what goes wrong with X" questions.

Inputs: query (required), limit, track.

search_research_papersA

Search 1,600+ real academic papers underpinning the research (title, authors, abstract, DOI). Use when a claim needs a citable primary source.

Inputs: query (required), limit.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.4/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool targets a distinct content type: FAQ entries, pain points, structural drivers, jurisdiction statuses, research papers, and corpus metadata. The search tools are differentiated by their object type, and the descriptions explicitly state when to use each, so there is no overlap.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern (e.g., search_faq, search_pain_points, list_structural_drivers, lookup_jurisdiction, search_research_papers, get_corpus_stats). Verbs are action-appropriate and nouns clearly indicate the resource.

    Tool Count5/5

    With 6 tools, the server is well-scoped for its purpose of accessing a privacy research corpus. Each tool covers a distinct content type or metadata function, and none are redundant.

    Completeness5/5

    The server covers discovery and search across all major corpus components (FAQ, pain points, structural drivers, jurisdictions, research papers) plus a stats/metadata tool for valid filters. There are no obvious dead ends, as every tool returns the information needed for its use case.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      Zero-auth multi-source research MCP server that enables web search, reading URLs, PDFs, GitHub repos, and querying Hacker News, Stack Overflow, Semantic Scholar, and YouTube transcripts without API keys.
      58 PyPI
      12
      Apache 2.0
    • A
      license
      Not graded
      quality
      D
      maintenance
      MCP server for grounded agentic Q&A over customer feedback, exposing typed tools to query a feedback corpus and return answers with citations to specific record IDs or a refusal when unsupported.
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      A Personal Knowledge-Base MCP Server that provides semantic search over a student-owned document collection using MCP, Gemini embeddings, and Qdrant. It exposes tools for searching notes, retrieving full documents, and listing indexed sources.
      -
    • A
      license
      Not graded
      quality
      C
      maintenance
      Read-only MCP server that exposes the curated Sumi Docs corpus through four tools, enabling agents to query documentation content programmatically.
      1
      MIT