MyTown MCP server

An MCP server that gives an AI assistant grounded, source-linked access to US & Canadian local-government meetings — what your city council, county board, or school board has coming up and recently decided, in plain English, plus place context (demographics, home-value trends, federal awards, permits, campaign finance, EPA/nonprofit ties). Every record links back to the official primary source.

Data comes from MyTown's public, keyless HTTP API, so this server needs no API key and no local database — it works out of the box for anyone. Coverage: ~7,000 places (6,616 US + 428 Canadian jurisdictions publishing meetings), 2.06M meeting records, refreshed daily. Coverage is not uniform — it follows what each government publishes; see /coverage/ for every place and every gap.

Tools

Per-city (great for "what's happening in my town"):

Search (new in 0.1.2):

Whole-dataset (cross-city — every table, no download; backed by HuggingFace's hosted dataset-viewer API):

Typical flows: list_cities("palo alto") → get_city_context(slug) for one place; or dataset_tables() → search_dataset("police budget", "decisions") to sweep the whole country.

What's in the corpus

Beyond meetings and AI briefs, the published dataset carries layers most local-government tools do not, each matched to a municipality and each linked to its primary source:

• fac_audits / fac_findings — the Federal Audit Clearinghouse local-government universe, 170,366 single audits and 50,626 findings including material weaknesses and repeat findings with the auditor's verbatim narrative
• member_votes / people — per-official roll calls, and the officials who cast them
• lobbying_registrations / lobbying_expenditures / official_disclosures — who is paid to influence local government, and what officials disclosed
• civic_records — permits, licences and code cases where the jurisdiction publishes them
• muni_financials / debt_issues / gov_payments — budgets, bond issues, vendor payments
• transcripts — meeting-video text where published

Call dataset_tables() for the authoritative live list.

Provenance, for anyone citing this

Every brief records which model wrote it (briefs.model) and when (briefs.created_at). The prompts are published verbatim and versioned by content hash at /methodology/. Summaries are AI-generated from official documents — always verify against the linked primary source. Figures and vote tallies are covered by a numeric grounding audit; wording and emphasis are not.

Install & configure

Claude Desktop

Add to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "mytown": {
      "command": "uvx",
      "args": ["mytown-mcp"]
    }
  }
}

Or run from a local checkout (no publish needed):

{
  "mcpServers": {
    "mytown": {
      "command": "/absolute/path/to/mcp-server/.venv/bin/python",
      "args": ["-m", "mytown_mcp.server"]
    }
  }
}

Claude Code

# from a local checkout:
claude mcp add mytown -- /absolute/path/to/mcp-server/.venv/bin/python -m mytown_mcp.server
# or, once published:
claude mcp add mytown -- uvx mytown-mcp

From source

python -m venv .venv && .venv/bin/pip install mcp httpx
.venv/bin/python -m mytown_mcp.server   # speaks MCP over stdio

Example questions it can answer

• "What's the Palo Alto city council deciding this month?"
• "Show recent school-board meetings for the Palo Alto Unified School District."
• "For Lake Forest, CA — pull demographics, home-value trend, and any federal awards."
• "Across the whole country, which towns passed rent-control decisions this year?" (search_dataset)
• "Show the largest federal awards in the dataset, over $1B." (filter_dataset)
• "Find meetings tied to EPA facilities with an identified violation." (filter_dataset on decision_echo)

Notes

• Meeting/decision summaries are AI-generated from official documents — always verify against the linked primary source before relying on a detail.
• License: CC BY 4.0 (attribution: mytown.theboringparts.com).
• For research spanning many cities (training, full-text search, bulk stats), use the open dataset rather than looping these tools.