AI GitHub
mcp-agent for workflows around MCP servers
Compose agent workflows that discover tools through MCP connections.
Key Takeaways
- The library combines model-backed workflows with managed MCP server connections.
- Decide which servers the workflow can connect to and what model provider will run it.
- Connect one read-only test server and run a workflow over a known fixture.
- More servers expand the tool choices and permission surface.
How it works
The library combines model-backed workflows with managed MCP server connections. Workflow patterns let applications coordinate tool calls and hand results between agents.
What you need
Decide which servers the workflow can connect to and what model provider will run it. Use the tool discovery result to verify the capabilities exposed to each step.
What to try first
Connect one read-only test server and run a workflow over a known fixture. Save the requested tool names and arguments so you can compare runs without relying on the final text.
Use two tools whose descriptions make their scope distinct. Keep a record of the routes chosen by the workflow when the same request is repeated.
Where to look for failures
More servers expand the tool choices and permission surface. Test capability collisions and reduce the tools visible to each workflow.
Related Franklin pages
Use one read-only server to separate connection behavior from agent choices. A deterministic tool result makes it easier to trace failures through a larger workflow.
- Use higher-level client and server tooling for MCP-enabled applications. mcp-use for connecting applications to MCP tools
- Build Python agents with typed dependencies, tools and model outputs. Pydantic AI for typed agent applications
Browse Franklin for concrete examples of tools and services exposed through MCP. Franklin’s MCP directory
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗