PubRecords MCP

Public-records intelligence for AI agents. Aggregate US business filings, SEC reports, federal court cases, federal spending awards, and professional licenses through a single MCP server. Built for due diligence, compliance, sales research, and background checks.

MCPize Listing Copy

PubRecords MCP turns the patchwork of US public-records APIs into a single, predictable tool surface for AI agents. One call cross-references a company across OpenCorporates, SEC EDGAR, USASpending.gov, CourtListener (RECAP), the NPI registry, and FCC license-view — and returns a normalized confidence score so your agent doesn't need to babysit the messy bits. Every endpoint is async, retries upstream 429s with exponential backoff, and never crashes — failures come back as structured {success: false, error: ...} envelopes that downstream tools can branch on cleanly.

Designed for due diligence and KYB workflows, compliance and adverse-media checks, sales prospecting, and background research. AI agents using PubRecords MCP can verify whether a counterparty actually exists, find their officers and registered agent, surface federal litigation history, see contract awards from Uncle Sam, and confirm professional licenses — all without juggling six different API keys, rate-limit headers, or response schemas. Cached 24 hours by default; most underlying records refresh daily at the source. Free tier (30 calls/day) for evaluation; Pro at $29/mo or $0.03/call for production traffic.

Pricing

Default dev key (free tier): pubrecords-dev-key-001

Upgrade: https://mcpize.com/pubrecords-mcp

Tool Reference

Install in Claude

claude mcp add --transport http pubrecords-mcp https://mcp-pubrecords-production.up.railway.app/mcp/

Note the trailing slash on /mcp/. Without it the gateway issues a 307 that drops the POST body, breaking the JSON-RPC handshake.

Then set the API key in your client:

X-API-Key: pubrecords-dev-key-001

REST surface (mirrors the MCP tools)

All MCP tools are also available over plain HTTP for non-MCP clients:

curl -H "X-API-Key: pubrecords-dev-key-001" \
  "https://mcp-pubrecords-production.up.railway.app/tools/search_companies?name=Apple&state=CA"

Production URL: https://mcp-pubrecords-production.up.railway.app

GET  /health                          (no auth)
GET  /usage                           (X-API-Key)
GET  /tools/<tool_name>?...           (X-API-Key)
POST /mcp/                            (MCP streamable-http transport)

GET /usage returns your remaining daily quota. GET /health is unauthenticated.

Local development

git clone https://github.com/bch1212/mcp-pubrecords.git
cd mcp-pubrecords
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn server:app --reload --port 8000

Run tests:

pytest -v

Optional environment variables (see .env.example):

• PUBRECORDS_SEC_USER_AGENT — SEC EDGAR requires an identifying UA. Set it in production.
• OPENCORPORATES_TOKEN — raises the OpenCorporates anonymous cap of ~50/month.
• COURTLISTENER_TOKEN — raises CourtListener's anonymous rate limit.
• PUBRECORDS_CACHE_TTL — cache TTL in seconds (default 86400 = 24h).
• PUBRECORDS_ADMIN_TOKEN — required to issue new keys at /admin/keys.

Data freshness

Responses are cached for 24 hours by default. Underlying source agencies update on different cadences:

For ultra-fresh queries, set PUBRECORDS_CACHE_TTL=300 (5 minutes) — note that this will multiply your upstream API costs.

Data Gaps & Source Limitations

These are honest caveats — the MCP returns structured not_implemented envelopes rather than fabricated data when a query falls outside coverage:

1. UCC filings — there is no uniform federal UCC API. Most state Secretary-of-State systems are HTML-only or require per-state credentials. get_ucc_filings returns {"error": "not_implemented"} until per-state adapters land (CA, DE, NY, TX are the priority backlog).
2. State professional licenses — covered for healthcare (NPI) and FCC. Other categories (bar, real estate, contractor, CPA) are state-by-state with no central API. search_licenses will say which sources matched.
3. OpenCorporates rate limits — anonymous traffic is capped at ~50 requests/month per IP. Set OPENCORPORATES_TOKEN for production. The 24h cache absorbs most of the burn.
4. SEC EDGAR full-text — search_sec_filings searches the body of filings, not just metadata, so query terms common across many filings can return wide result sets. Combine with form_type to narrow.
5. CourtListener coverage — federal only (PACER/RECAP). State court coverage is uneven and not guaranteed.
6. No PII / consumer-record sources — by design. PubRecords is for entity-level due diligence; consumer credit and identity data are deliberately out of scope.

License & Compliance

This MCP only proxies and caches public records that are already free to access from federal sources. Cached payloads are kept for 24 hours and can be flushed by deleting the SQLite DB. No PII is stored beyond what the underlying public-record agencies already publish.