MCP server for agent spend
Five read-only tools for an agent
An agent does not run a shell. The MCP server exposes the same checks as the CLI over JSON-RPC stdio. Input is data only: no file paths, no network, no chain access, no keys.
verify_receipt— Proof of Spend checksaudit_spend— CSV rows → findingsreserve_report— RWA NAV / proof of reserveresolve_report— prediction evidencereconcile_books— invoices ↔ transfers
Run it
python3 -m nidhipati_proof.mcp_stdio # or docker build -t nidhipati-mcp . docker run -i --rm nidhipati-mcp
Read-only tools. A failed check returns isError with the report, not a protocol failure.
Client config
{
"mcpServers": {
"nidhipati": {
"command": "python3",
"args": ["-m", "nidhipati_proof.mcp_stdio"]
}
}
}
No API key, no account, no outbound network. The agent sends data; the server returns a report.
Official MCP Registry: io.github.mrpkk/nidhipati · Glama: glama.ai/mcp/servers/mrpkk/nidhipati · OCI: ghcr.io/mrpkk/nidhipati:0.6.0
Why MCP
Catalog discovery is the one channel that already produced measured traffic for our attest server. MCP lets an agent call the audit without a shell — and lets a catalog list it. The server is intentionally read-only: it cannot move money.
Verify a Proof of Spend receipt in your browser — Ed25519, no upload.