MCP Integration

Connect external tools via the Model Context Protocol (MCP).

What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI models to external tools and data sources. YokeBot supports MCP as a way to give your agents access to tools hosted on external MCP servers without writing custom skill files.

How MCP Works in YokeBot

When you configure an MCP server connection, YokeBot queries the server for its tool manifest, which lists available tools and their parameter schemas. These tools then appear in the skill picker and can be assigned to agents just like built-in skills.

Configuring an MCP Server

From the dashboard, go to Settings > MCP Servers and click "Add Server". Provide:

Field	Description
Name	A display name for this server connection.
URL	The MCP server endpoint URL.
API Key	Authentication key if the server requires one.
Auto-Sync	Whether to re-fetch the tool manifest periodically.

For self-hosted instances, you can also configure MCP servers via environment variables:

MCP_SERVERS='[
  {
    "name": "my-tools",
    "url": "https://mcp.example.com",
    "apiKey": "sk-..."
  }
]'

Assigning MCP Tools to Agents

Once an MCP server is connected and its tools are synced, the tools appear in the skill picker when editing an agent. MCP tools are prefixed with the server name to distinguish them from built-in skills (e.g., "my-tools:send_email").

Security Considerations

warning

MCP tools execute on an external server you may or may not control. Only connect to MCP servers you trust. Review the tool manifest carefully before assigning tools to agents, especially tools that modify external systems.

Troubleshooting

If tools do not appear after adding a server, check that the URL is reachable and the API key is valid.
Enable debug logging in the engine to see raw MCP handshake and tool discovery responses.
Ensure the MCP server implements the standard tool manifest endpoint.