/ developers · mcp
Give your agent the integration layer.
@inferrex/mcp exposes Inferrex as Model Context Protocol tools, so Claude, Cursor and other agents can connect an API, infer its schema, and read governed data — with your guardrails, not their guesswork.
Connect it to Claude Desktop / Cursor
The server runs locally over stdio and talks to the API with your key. Add it to your MCP client config:
{
"mcpServers": {
"inferrex": {
"command": "npx",
"args": ["@inferrex/mcp", "--api-key=YOUR_INFERREX_API_KEY"]
}
}
}Or run it directly:
INFERREX_API_KEY=your_key npx @inferrex/mcp
# or: npx @inferrex/mcp --api-key=your_key --api-url=https://api.inferrex.comThe @inferrex/mcp package and the hosted mcp.inferrex.com endpoint both arrive at launch — the setup above and the tools below are the surface you'll get. The REST API and your API keys are live today.
The tools
Each tool is one API operation, named in snake_case, and requires your API key. A sample of what your agent gets:
| Tool | Does |
|---|---|
| `connection_infer` | Point Inferrex at an API URL and infer its schema |
| `connection_status` / `connection_manifest` | Check inference progress and read the classified schema |
| `connection_approve` | Approve an inferred manifest |
| `pipeline_create` / `pipeline_sync` | Wire a source→target pipeline and run a sync |
| `pipeline_suggestions` | Read self-healing repair suggestions |
| `library_search` | Search the connector corpus for a provider |
| `intelligence_usage` | Read workspace usage meters |
The full tool set mirrors the REST API one-to-one — anything on the API reference is reachable as a tool. Tools that touch a gated capability surface a clear "upgrade tier" message rather than failing silently.
A typical agent flow
"Connect our Stripe account and tell me which fields are payment-sensitive."
The agent calls connection_infer with the Stripe URL, polls connection_status, reads connection_manifest, and reports back the fields Inferrex flagged as sensitive — without you writing a line of mapping code.