@conovo/mcp

An MCP server that lets a coding agent integrate Conovo into your platform — and debug the integration afterwards.

Point Claude Code (or any MCP client) at it and say "integrate Conovo": the agent can read your data model, build and register your payload schema, mint sandbox sessions, drive test contracts through the real API, wire up webhooks, and read the request inspector when something 4xxes.

Setup

claude mcp add conovo -e CONOVO_SECRET_KEY=sk_test_… -- npx -y @conovo/mcp

Or in any MCP client config:

{
  "mcpServers": {
    "conovo": {
      "command": "npx",
      "args": ["-y", "@conovo/mcp"],
      "env": { "CONOVO_SECRET_KEY": "sk_test_…" }
    }
  }
}

Use an sk_test_ key while integrating — every session it mints is test mode end to end: simulated signing, nothing metered, nothing binds. CONOVO_API_URL overrides the API base (defaults to https://api.conovo.co).

Tools

| Tool | What it does | |---|---| | get_overview | Account status + live contract metrics | | get_payload_schema / register_payload_schema | Read / append the payload sample contract fields bind to | | infer_payload_schema | Propose a schema from a sample response, OpenAPI spec, or Prisma schema | | get_payload_gaps | Which fields keep missing from the payload — the schema backlog | | create_sandbox_session | A real /v1 token against the Sandbox workspace | | get_api_reference | Routes + summaries from the live OpenAPI doc | | list_requests | The request inspector — why a call failed, no bodies ever | | get_held_contracts | Contracts held by validation, with issue codes | | list_events | The content-free audit trail | | get_webhook / set_webhook / test_webhook | Outbound webhook config + a signed test ping |

What it deliberately cannot do

No tool touches a live contract, produces a document value, or returns contract content. The send path stays deterministic and human-gated — the agent integrates the product; it doesn't operate it.

License

MIT