microquery-mcp

MCP server for Microquery — ask research questions about real-world data and get actual database records back. Works in Claude Desktop, Cursor, or any MCP-compatible AI host. Claude handles the SQL; you just ask the question.

No wallet required. Auto-registers on first query and runs on $0.10 trial credit (~1,600 typical queries).

See also

microquery-agent — autonomous agent for pipelines and cron jobs: register → deposit USDC → query → auto top-up. Use this if you want to run microquery unattended without an AI host.

Prerequisites

• Python 3.9+
• No third-party packages — stdlib only

Installation

Claude Desktop (recommended)

The easiest paths, in order of friction:

1. Registry install Find microquery in the Claude Desktop MCP marketplace or on smithery.ai and click Install. No config editing needed.

2. uvx install Add one entry to ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

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

uvx requires uv to be installed separately (astral.sh/uv). Once installed, Claude Desktop will pick it up automatically.

3. Manual install (developers) Download microquery_mcp.py from this repo and point your host at it:

{
  "mcpServers": {
    "microquery": {
      "command": "python3",
      "args": ["/path/to/microquery_mcp.py"]
    }
  }
}

Restart Claude Desktop after editing. The server appears in Settings → Developer with a green dot when connected.

Other MCP hosts (Cursor, etc.)

Use the same command / args pattern above. Consult your host's MCP documentation for the exact config format.

Tools

How it works

1. On the first query() call the server registers an account using your OS username (getpass.getuser()) via POST /v1/register and stores the API key in ~/.microquery/token.
2. Subsequent calls use the stored key — no configuration needed.
3. The new account starts with 100,000 µUSDC ($0.10) trial credit, covering roughly 1,600 typical queries.
4. When trial credit runs low the server automatically tops up the account (up to 10 times, $2 each). Once the free allowance is exhausted a checkout URL is returned — visit it to add USDC and continue querying.

Available datasets

FDA adverse events · SEC EDGAR · clinical trials · ClinVar · arXiv · PubMed · Ethereum · Bitcoin · Base · DeFi TVL · FEC contributions · FRED economic series · NVD/CVE · OSV advisories · sanctions · FHFA house prices · GWAS · ClinPGx · malware samples · open food facts · world bank commodities · and more — call list_databases() for the full live schema.

Example

User:    What were the top adverse events reported for metformin last year?
         And how does that compare to 2022 and 2023?
Claude:  [queries fda.faers for each year, builds trend table]
         GI events (diarrhoea, nausea, vomiting) were flat 2022→2024,
         then spiked sharply in 2025 — consistent with the longevity/
         obesity wave hitting FAERS with a lag. Lactic acidosis stayed
         nearly flat across all four years despite overall volume growth.

User:    Can you cross-reference that with genomic profiles?
Claude:  [queries clinpgx, clinvar, gwas — no SQL needed from user]
         SLC22A1 rs628031 has a direct ClinPGx annotation for GI toxicity —
         the strongest known genomic explanation for why diarrhoea and nausea
         dominate the FAERS signal for metformin.