MCP Server Integration (optional)

Overview

MCP (Model Context Protocol) servers expose tools (e.g. search, APIs) that an agent can call. If Agent OS supports MCP in your build, each server’s tools are typically registered with a prefix (e.g. mcp_<name>_<tool>) so they don’t clash with built-in or skill tools.

• Config: May be in dashboard (e.g. Settings or Integrations → MCP Servers), or in a config file (e.g. ~/.agent-os/mcp.json), or via env (e.g. MCP_SERVERS).
• Discovery: Community servers are listed at mcpservers.org.

If supported in your build

1. Add one or more MCP servers (name, command, args, optional env for API keys).
2. On startup (or after saving config), the runtime may spawn each server over stdio, list its tools, and register them as agent tools.
3. The model sees these tools with the rest of the tool list; when it calls one, the runtime forwards the call to the MCP server and returns the result.

Check the Agent OS repo for:

• Whether MCP is implemented (e.g. in core/plugins.ts or a dedicated MCP module).
• Exact config location and format (mcp.json, env, or dashboard-only).
• How tool names are prefixed and how they are exposed to agents.

If MCP is not yet implemented, this page serves as a placeholder for when it is added. Until then, extend the agent with skills (tools.yaml) and built-in tools (see Skills authoring and Agent OS).