Cargo MCP server

The Cargo MCP server is how Claude, ChatGPT, Cursor, and any other MCP client call a workspace: a curated set of your tools, agents, and data models behind one Model Context Protocol endpoint.

You define a server with defineMcpServer. Cargo hosts it over Streamable HTTP with OAuth, or you serve it locally over stdio with cargo-ai mcp.

Hosted HTTP

https://mcp.getcargo.io/v1/ai/mcpServers/<uuid>/mcp

Hosted clients (ChatGPT custom connectors, Claude.ai, Cursor over HTTP) connect to that URL and sign in with OAuth. There is no token to paste. The discovery URL https://mcp.getcargo.io/mcp answers 401 with a WWW-Authenticate challenge rather than 404; the token you obtain resolves to your own workspace.

Server card

The MCP server card — name, icon, description, transport — is at:

• https://www.getcargo.ai/.well-known/mcp/server-card.json
• https://mcp.getcargo.io/.well-known/mcp/server-card.json

/.well-known/mcp and /.well-known/mcp.json on this host proxy the same card.

stdio, from a coding agent

claude mcp add cargo -- cargo-ai mcp --server <uuid>

cargo-ai mcp uses the credentials the CLI already holds. With no --server, it falls back to CARGO_MCP_SERVER_UUID, or to the workspace’s only MCP server when there is exactly one.

Define a server

import { defineMcpServer } from "@cargo-ai/cdk";

export const crmServer = defineMcpServer("crm", {
  description: "CRM tools and data for assistants.",
  uses: [enrich, sdr, { ref: contacts, readOnly: true }],
});

Deploy with cargo-ai cdk deploy, then cargo-ai ai mcp-server list to read the uuid.

Docs MCP

Documentation has its own public MCP server at https://docs.getcargo.ai/mcp, for search and retrieval. The product server above is the one that acts on a workspace.

Related

• Overview and client setup: https://docs.getcargo.ai/mcp-servers/overview
• Cargo auth docs
• Cargo API docs