Connect jpcite to Claude Code in 30 seconds

Five-minute setup

1. Install uvx: run curl -LsSf https://astral.sh/uv/install.sh | sh and restart your shell.
2. Register jpcite as an MCP server: claude mcp add jpcite -- uvx autonomath-mcp (the PyPI name stays autonomath-mcp for legacy reasons; the brand is jpcite).
3. Verify the connection: success means claude mcp list | grep jpcite shows up in green.
4. (Optional) Set an API key: to remove the 3 req/day cap, run export JPCITE_API_KEY=<YOUR_JPCITE_API_KEY>. Issue one from /en/pricing.html#api-paid.
5. Smoke test: confirm that 261 MCP tools are visible using the verification command in the next section.

Copy-paste snippet

To edit ~/.claude.json directly, add the following to the MCP servers block.

{
  "mcpServers": {
    "jpcite": {
      "command": "uvx",
      "args": ["autonomath-mcp"],
      "env": {
        "JPCITE_API_KEY": "<YOUR_JPCITE_API_KEY>"
      }
    }
  }
}

If you do not need an API key (anonymous free, 3 req/IP/day), you can omit the env block entirely.

Verification

Run the /mcp command inside a Claude Code session (or the CLI below). Success means jpcite shows a green dot and its 261 tools are loaded.

claude mcp list   # jpcite shows "Connected"

Expected: jpcite: uvx autonomath-mcp - ✓ Connected appears and the 261 published tools are loaded. Generation of 36-Agreement labor documents is not offered publicly.

When smoke-testing the REST endpoint directly, always use -G --data-urlencode. Embedding a Japanese query straight into the URL breaks the HTTP request line, and the API returns {"error":{"code":"bad_request",...,"hint":"Use --data-urlencode for non-ASCII query params..."}}.

curl -G "https://api.jpcite.com/v1/programs/search" \
  --data-urlencode "q=補助金" \
  --data-urlencode "prefecture=東京都" \
  -H "X-API-Key: ${JPCITE_API_KEY:-}"

Before stuffing long PDFs or search results into every claude -p or Claude Agent SDK prompt, build a jpcite packet first so you can inspect jpcite_cost_jpy, estimated_tokens_saved, source_count, and known_gaps. This does not guarantee a reduction in your Claude bill, but for repeated work it lets you design around passing a small, evidence-backed input.

packet="$(curl -sS -X POST "https://api.jpcite.com/v1/evidence/packets/query" \
  -H "X-API-Key: ${JPCITE_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{"query_text":"東京都 製造業 設備投資 補助金","limit":5,"include_compression":true,"source_tokens_basis":"pdf_pages","source_pdf_pages":30,"input_token_price_jpy_per_1m":300}')"

echo "$packet" | jq '{jpcite_cost_jpy, estimated_tokens_saved, source_count, known_gaps}'
claude -p "Using only this jpcite Evidence Packet as your basis, organize the candidate programs, clarifying questions, and known_gaps. Do not assert anything as professional judgment. $packet"

FAQ