📝Cargo exposes workspace tools, agents, data models, and connector actions as 📝Model Context Protocol (MCP) servers — making the full GTM automation layer callable from any MCP client (Claude Code, 📝Claude Desktop, Cursor, etc.).
Cargo's MCP works bidirectionally. As a server, it exposes a workspace's tools, agents, and data to external AI clients. As a client, it connects external MCP servers into Cargo agents via the connect-mcp-client API endpoint. What gets exposed falls into two categories: actions (tools, agents, and connector integrations like HubSpot or Salesforce) and resources (data models with optional filtering and files stored in the workspace).
Two server shapes exist. Every workspace has a platform server at https://mcp.getcargo.io/mcp that exposes the whole action catalogue, models, and context through tools such as search_actions, execute_action, execute_action_batch, and query_models. Curated servers, created in the UI, by API, by the 📝Cargo CLI, or by defineMcpServer in the 📝Cargo CDK, expose a chosen subset at https://mcp.getcargo.io/v1/ai/mcpServers/<uuid>/mcp. Both speak Streamable HTTP and accept either an OAuth grant or a workspace API token as a bearer header. The CLI's cargo-ai mcp command bridges any Cargo server to local stdio using the CLI's own credential, so a client like Claude Code needs no token in its config; mcp-remote remains the fallback for clients that cannot set headers.
Cargo is a core component of the 📝One Inc GTM engineering stack, providing the automation and enrichment layer that feeds into prospecting and outreach workflows.