fast-agent-mcp

What it does

fast-agent is a Python framework for building, testing, and deploying LLM agents with native MCP support. It provides a CLI-first interface with optional interactive terminal UI, letting you compose agents from simple declarative configuration files. The framework supports multiple LLM providers natively—Anthropic, OpenAI, Google, Azure, Ollama, Deepseek, and others—and includes features like structured outputs, vision and PDF support, MCP elicitations, and shell command execution within agent context.

Who it's for

AI engineers evaluating different models and MCP server combinations, developers building production agents with complex workflows, and teams that need rapid iteration between local development and cloud LLM providers. Useful for both interactive exploration and programmatic agent execution.

Common use cases

• Chat with agents interactively to diagnose and tune behavior during development without redeploying.
• Test the same agent against multiple LLM providers without code changes—swap between Opus, Sonnet, local Ollama models, etc.
• Connect and inspect MCP server interactions, including Streamable HTTP transport diagnostics for debugging reliability.
• Build evaluation platforms that measure agent performance across different models and configurations.
• Orchestrate multi-step workflows with human-in-the-loop approval gates and real-time agent inspection.

Setup pitfalls

• Eight secrets detected during scanning—handle LLM API keys, OAuth tokens, and other credentials securely; configure via environment variables or the built-in OAuth and KeyRing system.
• High risk classification due to filesystem and network access; isolate execution in sandboxed environments when running untrusted agents or connecting to remote MCP servers.
• Multiple LLM provider credentials may be needed for comparative testing; set these before launching agents to avoid runtime auth failures.
• Some quickstart commands require the uv package manager; pip install works but uv is recommended for seamless tool mode and dependency management.