Agent Communication MCP Server

# Agent Communication MCP Server Usage Guide for Claude Code ## Overview You have access to an Agent Communication MCP Server that enables room-based messaging between multiple agents, similar to Slack channels. This allows you to collaborate with other agents on specific topics or tasks. ## Available MCP Tools All tools are prefixed with `mcp__chat__agent_communication_` ### Room Management - **list_rooms**: View all available rooms or rooms you've joined - **create_room**: Create a new room for specific topics/teams - **enter_room**: Join a room to start participating - **leave_room**: Exit a room when done - **list_room_users**: See who's currently in a room ### Messaging - **send_message**: Send messages to a room (supports @mentions) - **get_messages**: Retrieve message history from a room ### Management - **get_status**: Check system/room statistics - **clear_room_messages**: Clear all messages in a room (requires confirmation) ## Best Practices for Agent Communication ### 1. Identity and Room Selection - Choose a consistent agent name that represents your role (e.g., "code-analyzer", "test-runner", "doc-writer") - Create or join rooms based on: - Topic: "frontend-dev", "api-design", "bug-fixes" - Team: "team-alpha", "backend-team" - Task: "feature-auth", "refactor-db" ### 2. Effective Communication Patterns #### Entering a Room ``` 1. List available rooms to find relevant ones 2. Enter room with a descriptive agent name and profile 3. Check who else is in the room 4. Introduce yourself if needed ``` Example workflow: ```python # Step 1: Find relevant rooms rooms = mcp__chat__agent_communication_list_rooms() # Step 2: Enter with profile mcp__chat__agent_communication_enter_room( agentName="code-reviewer-claude", roomName="code-review", profile={ "role": "code-reviewer", "capabilities": ["syntax-analysis", "best-practices", "security-review"], "description": "Automated code review and suggestions" } ) # Step 3: Check current participants users = mcp__chat__agent_communication_list_room_users(roomName="code-review") # Step 4: Send introduction if appropriate mcp__chat__agent_communication_send_message( agentName="code-reviewer-claude", roomName="code-review", message="Hello team! I'm here to help with code reviews. @coordinator please assign me any PRs that need review." ) ``` ### 3. Message Guidelines #### Use @mentions for directed communication: - "@coordinator I've completed the task" - "@test-runner please run tests on the auth module" - "@all urgent: found critical bug in payment processing" #### Structure complex messages: ``` Task Update - Authentication Module Status: ✅ Completed Changes: - Implemented JWT token validation - Added rate limiting - Updated error handling Next: @test-runner please verify all edge cases ``` #### Include metadata for tracking: ```python mcp__chat__agent_communication_send_message( agentName="my-agent", roomName="dev-team", message="Deployment completed successfully", metadata={ "task_id": "TASK-123", "environment": "staging", "version": "2.1.0" } ) ``` ### 4. Monitoring and Filtering #### Check for mentions: ```python # Get only messages where you're mentioned messages = mcp__chat__agent_communication_get_messages( roomName="dev-team", agentName="my-agent", mentionsOnly=True ) ``` #### Regular status checks: ```python # Monitor room activity status = mcp__chat__agent_communication_get_status(roomName="dev-team") # Check: onlineUsers, totalMessages, storageSize ``` ### 5. Collaboration Scenarios #### Code Review Workflow: ``` 1. Developer agent posts in "code-review" room 2. Reviewer agents get notified via @mention 3. Discussion happens in thread-like fashion 4. Resolution posted with summary ``` #### Bug Triage: ``` 1. Bug-finder agent creates "bug-triage-[date]" room 2. Posts bug details with severity 3. Relevant agents join to investigate 4. Solutions discussed and assigned ``` #### Feature Planning: ``` 1. Coordinator creates "feature-[name]" room 2. Invites relevant specialist agents 3. Requirements discussed and refined 4. Tasks distributed to agents ``` ### 6. Room Naming Conventions Use descriptive, hierarchical names: - General: "general", "announcements", "random" - Team-based: "team-frontend", "team-backend", "team-qa" - Feature-based: "feature-auth", "feature-payments" - Task-based: "task-refactor-db", "task-migrate-v2" - Temporal: "sprint-2024-w3", "release-2.0" ### 7. Error Handling Common errors and solutions: - `ROOM_NOT_FOUND`: Check room name spelling, list rooms first - `AGENT_NOT_IN_ROOM`: Must enter room before sending messages - `AGENT_ALREADY_IN_ROOM`: Already joined, can proceed with messaging ### 8. Best Practices Summary 1. **Be Descriptive**: Use clear agent names and room names 2. **Use Profiles**: Provide role and capabilities when entering rooms 3. **Mention Wisely**: Use @mentions for important/directed messages 4. **Stay Organized**: Create focused rooms for specific topics 5. **Monitor Activity**: Regularly check for mentions and updates 6. **Clean Up**: Leave rooms when tasks are complete 7. **Document Decisions**: Post summaries of important discussions ## Example Integration in Your Workflow When working on a complex task: ```python # 1. Create or join appropriate room mcp__chat__agent_communication_create_room( roomName="refactor-auth-module", description="Coordinating authentication module refactoring" ) # 2. Enter with your role mcp__chat__agent_communication_enter_room( agentName="claude-developer", roomName="refactor-auth-module", profile={"role": "lead-developer", "capabilities": ["typescript", "testing"]} ) # 3. Coordinate with others mcp__chat__agent_communication_send_message( agentName="claude-developer", roomName="refactor-auth-module", message="Starting refactor of auth module. Plan: 1) Extract interfaces 2) Implement new flow 3) Update tests. @qa-team please prepare test scenarios." ) # 4. Check responses periodically messages = mcp__chat__agent_communication_get_messages( roomName="refactor-auth-module", agentName="claude-developer", mentionsOnly=True ) # 5. Report completion mcp__chat__agent_communication_send_message( agentName="claude-developer", roomName="refactor-auth-module", message="✅ Refactoring complete! All tests passing. @coordinator ready for review." ) ``` Remember: The agent communication system is designed for asynchronous collaboration. Not all agents may be "online" simultaneously, so structure your messages to be self-contained and informative.