MCP Agents¶

Overview¶

Connect to external MCP servers and expose their tools as callable playbooks.

What is MCP?¶

The Model Context Protocol (MCP) is an open protocol that standardizes how AI applications can access external tools and data sources. MCP servers expose tools that can perform various tasks like:

Accessing databases
Making API calls
Performing calculations
Interacting with file systems
Running specialized algorithms
And much more

Create an MCP agent¶

To create an MCP agent, you need to specify the agent's remote configuration. The agent acts as a proxy to the MCP server, making all its tools available as playbooks.

Basic Example¶

# My MCP
This agent provides various python tools.
remote:
  type: mcp
  url: http://127.0.0.1:8888/mcp
  transport: streamable-http

Once defined, you can call tools from the MCP server in your instructions:

# Assistant
## Main
### Triggers
- When program starts

### Steps
- Call ListProjects from My MCP
- Display the project list to the user
- End program

Configuration Options¶

The MCP agent configuration supports the following options:

Option	Required	Description	Default
`type`	Yes	Must be `mcp` for MCP agents	-
`url`	Yes	The MCP server URL or command	-
`transport`	No	Transport type (`sse`, `stdio`, `websocket`, `streamable-http`)	`sse`
`timeout`	No	Timeout in seconds for tool calls	`30.0`
`auth`	No	Authentication configuration object	`{}`

Transport types¶

MCP agents support different transport protocols:

SSE (Server-Sent Events): Default transport for HTTP-based MCP servers
streamable-http: HTTP-based streaming transport
stdio: Standard input/output for local processes
WebSocket: WebSocket-based transport for persistent connections

How MCP Agents Work¶

Connection: When the program starts, the MCP agent connects to the specified server
Tool Discovery: The agent automatically discovers all available tools from the server
Playbook Creation: Each MCP tool becomes a playbook that can be called
Execution: When a playbook is called, the agent forwards the request to the MCP server

Use MCP tools as playbooks¶

Once connected, all MCP tools are automatically available as playbooks. You can call them just like any other playbook:

# Weather MCP
This agent connects to a weather MCP service.
remote:
  type: mcp
  url: http://weather-service.example.com/mcp
  transport: sse

# Weather Assistant
## Main
### Triggers
- When program starts

### Steps
- Say "Welcome to Weather Assistant! What location would you like weather for?"
- Get user's location
- Call get_current_weather from Weather MCP with location=$location
- Display the weather: "Current weather in $location: $weather"
- End program

Combine multiple MCP agents¶

You can use multiple MCP agents in a single program:

# Weather MCP
Weather service integration.
remote:
  type: mcp
  url: http://weather-api.com/mcp

# News MCP
News service integration.
remote:
  type: mcp
  url: http://news-api.com/mcp

# Assistant
## DailyBriefing
### Steps
- Call get_current_weather from Weather MCP for New York
- Call get_top_headlines from News MCP with technology category
- Create briefing combining weather and news
- Present briefing to user

Authentication¶

For MCP servers that require authentication:

# Secure API Agent
Connects to a secured MCP endpoint.
remote:
  type: mcp
  url: https://api.example.com/mcp
  transport: sse
  auth:
    type: bearer
    token: ${API_TOKEN}

Run an MCP server (examples)¶

You can create simple MCP servers with fastmcp. Example servers in the repo:

Insomnia server: tests/data/insomnia/mcp.py
Travel advisor server: tests/data/travel_advisor/mcp.py