DEVELOPER REFERENCE / MCP

Tools for your agent.

The current MCP server runs locally over stdio. It is separate from the hosted HTTP inference API.

01 / Prepare the local server

Use an authorized checkout of the backend repository. In its omnideploy directory, configure the local .env with DATABASE_URL and provider credentials, install dependencies with pnpm install --frozen-lockfile, run the database migration, and build with pnpm build. Use an API key provisioned against that same database. The local MCP process needs database and provider access; a hosted router key alone is not sufficient.

02 / Configure your client

Add the following stdio server to your MCP client configuration. Replace the absolute path and key. Ensure Node is available to the client, or use an absolute Node executable path. Restart the client after saving.

{
  "mcpServers": {
    "omnideploy": {
      "command": "node",
      "args": [
        "/absolute/path/to/omnideploy/dist/routes/mcp.js"
      ],
      "env": {
        "OMNIDEPLOY_API_KEY": "YOUR_PROVISIONED_KEY"
      }
    }
  }
}

03 / Inspect available tools

route_inference

Run a supported inference request and record the provider attempt.

get_pricing_snapshot

Read configured pricing estimates. This is not live provider billing.

list_providers

List providers and models configured for this process.

The stdio transport is covered by automated integration tests. A live conversation inside each desktop client has not been verified. The current router does not expose an HTTP /mcp endpoint.

Use the hosted HTTP API instead →