Multi-Agent AI Collaboration with MCP is a lesson in the GenAI Unplugged MCP Masterclass that shows how to run two Claude Desktop conversations, a Researcher agent and a Writer agent, connected to the same MCP server so one saves research findings and the other reads them to write a polished draft.
What you will be able to do
- Build an MCP server named collaboration-hub with three tools: save_research, read_research, and save_draft
- Run two separate Claude Desktop conversations as a Researcher agent and a Writer agent that share work through one MCP server
- Decide when a task calls for a single agent versus multiple specialized agents, using the decision framework from the lesson
- Update your Claude Desktop config to add a new MCP server alongside an existing one, using your Python path and server.py path
- Verify your MCP SDK installation with check_setup.py before writing the server code
Before you start
- Completed the previous lesson's MCP server build, or have run pip install mcp to get the SDK
- Python 3 installed, with the ability to create and activate a virtual environment
- Claude Desktop installed and signed in (from claude.ai/download), used as the MCP host
- A code editor such as VS Code, Cursor, or PyCharm
Reference
| Step / Tool | Purpose | Details |
|---|---|---|
| check_setup.py | Verify MCP SDK install | Imports mcp.server.Server, stdio_server, Tool, TextContent; run with python check_setup.py |
| pip install mcp | Install the MCP SDK | Skip this step if you already installed it in the previous lesson |
| save_research tool | Save research findings | Writes content to research_findings.txt and confirms it saved |
| read_research tool | Read saved research | Reads research_findings.txt; reports an error if the file does not exist yet |
| save_draft tool | Save the final document | Writes content to final_draft.txt and confirms it saved |
| server.py | Defines the MCP server | Server named collaboration-hub; registers the three tools via handle_list_tools and handle_call_tool |
| Claude Desktop config | Connect Claude to the new server | Settings, Developer tab, Edit Config; add collaboration-hub alongside the existing note-reader server using your Python path and server.py path |
Common errors and fixes
| What goes wrong | The fix |
|---|---|
| MCP SDK import fails when running check_setup.py | The script prints the ImportError and tells you to run pip install mcp |
| read_research is called before any research has been saved | The tool returns: No research findings found. The Researcher agent needs to save research first using the save_research tool |
| save_research or save_draft is called with no content | The handler returns: Error: No content provided to save |
| A file read or write fails for another reason | The handler catches the exception and returns a message like Error saving research: (the exception text), or Error reading research: (the exception text) |
Read the full walkthrough
The complete lesson, with screenshots and any downloads, is published on Substack as part of MCP Masterclass: Connect AI to Everything.
More in this section
- Lesson 4: The Three Superpowers of MCP - Tools, Resources, and Prompts
- Lesson 5: Build an MCP Server in 30 Minutes
Continue the course
Browse all lessons in the MCP Masterclass: Connect AI to Everything course, or subscribe to the GenAI Unplugged newsletter to get new lessons in your inbox.
