Mcp Server

What is Mcp Server?

Mcp Server is a framework for creating and interacting with sophisticated AI Agents and Workflows, with complete, end-to‑tested support for the Model Context Protocol (MCP) including Sampling. It supports both Anthropic and OpenAI models, is multi‑modal (images and PDFs), and uses a simple declarative syntax stored in configuration files.

How to use Mcp Server?

Install via uv pip install fast-agent-mcp, then run uv run fast-agent setup to create example config files. Define agents with Python decorators like @fast.agent and invoke them with await agent("message") or open an interactive chat with await agent.interactive(). Use the --model flag to specify a model (e.g., uv run sizer.py --model sonnet).

Key features of Mcp Server

• First framework with end‑to‑end MCP Sampling support.
• Supports Anthropic (Haiku, Sonnet, Opus) and OpenAI models.
• Multi‑modal: handles images and PDFs via prompts, resources, and MCP tools.
• Declarative agent and workflow definitions in simple files.
• Built‑in workflow patterns: Chain, Parallel, Evaluator‑Optimizer, Router, Orchestrator.
• Agents can request human input for additional context.

Use cases of Mcp Server

• Building a chain of agents that fetch and summarize content from the web.
• Creating a multi‑language translator using parallel agent execution.
• Generating and refining reports with an Evaluator‑Optimizer loop.
• Routing complex user requests to the most capable agent using an LLM‑driven router.
• Orchestrating multiple agents to break down and complete large tasks.

FAQ from Mcp Server

What dependencies does Mcp Server require?

Mcp Server requires Python and the uv package manager. Install the package with uv pip install fast-agent-mcp.

What models can be used with Mcp Server?

Both Anthropic models (Haiku, Sonnet, Opus) and OpenAI models (gpt‑4o/gpt‑4.1 family, o1/o3 family) are supported. Models can be specified per agent or via the --model command‑line flag.

Where does Mcp Server store agent configurations?

Prompts and configurations are stored in simple files (e.g., fastagent.config.yaml). The framework looks for these files in the current directory and then recursively in parent directories.

Can Mcp Server use external MCP servers?

Yes, agents can be connected to MCP servers by listing their names in the servers parameter of the agent decorator. Servers must be defined in the configuration file.

Does Mcp Server support multimodal input?

Yes, both Anthropic and OpenAI endpoints support images and PDFs via prompts, resources, and MCP tool call results.