MCP Directory

Agent-MCP

Multi-agent collaboration protocol: orchestrate specialized AI agents over MCP with a shared knowledge graph.

Unverified
stdio (local)
API key
Python
View repo 1.2k npm

Add to your client

Copy the config for your MCP client and paste it into its config file.

Install / run
uv venv && uv install

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "agent-mcp": {
      "command": "uv",
      "args": [
        "run",
        "-m",
        "agent_mcp.cli",
        "--port",
        "8080"
      ],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key"
      }
    }
  }
}

Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf

Before you start

  • Python 3.10+ (with uv or pip)
  • Node.js 18.0.0+ (recommended 22.16.0) and npm 9.0.0+ for the dashboard / Node implementation
  • OpenAI API key (for embeddings and RAG)
  • 4GB RAM minimum
  • An AI coding assistant such as Claude Code or Cursor

About Agent-MCP

Agent-MCP turns a single AI assistant into a coordinated team of specialized, short-lived agents. An admin agent loads a Main Context Document (MCD) into a shared, RAG-searchable knowledge graph, then spawns purpose-built worker agents that execute granular linear tasks in parallel. File-level locking prevents agents from overwriting each other's work, and a real-time dashboard visualizes agents, tasks, and memory health. The framework runs as an MCP server, exposing its agent-management, task-orchestration, knowledge, and communication tools to MCP clients such as Claude Desktop.

Tools & capabilities (12)

create_agent

Spawn specialized agents (backend, frontend, testing, etc.).

list_agents

View all active agents and their status.

terminate_agent

Safely shut down agents.

assign_task

Delegate work to specific agents.

view_tasks

Monitor task progress and dependencies.

update_task_status

Track task completion and blockers.

ask_project_rag

Query the persistent knowledge graph.

update_project_context

Add architectural decisions and patterns to project context.

view_project_context

Access stored project information.

send_agent_message

Send a direct message between agents.

broadcast_message

Send an update to all agents.

request_assistance

Escalate complex issues.

When to use it

  • Coordinate multiple specialized AI agents working in parallel on different parts of a codebase
  • Maintain a persistent, searchable project knowledge graph so context is never lost between sessions
  • Decompose complex goals (e.g. building user authentication) into linear, atomic, parallelizable task chains
  • Prevent merge conflicts during AI-assisted development via automatic file-level locking
  • Monitor an AI development team in real time through a dashboard (agent status, task progress, memory health)

Security notes

Requires an OpenAI API key (set via OPENAI_API_KEY) for embeddings and RAG. The server uses admin/worker tokens; the admin token is printed in the server startup logs. Server can optionally bind to 0.0.0.0 and accept an --auth-token, so restrict network exposure accordingly.

Agent-MCP FAQ

Which implementation should I use?

The README recommends the Python implementation, started with `uv run -m agent_mcp.cli --port 8080`. A Node.js/TypeScript implementation is also available and can be installed globally via the agent-mcp-node npm package.

How do I connect it to Claude Desktop?

Add an entry to claude_desktop_config.json under mcpServers with command `uv` and args `["run", "-m", "agent_mcp.cli", "--port", "8080"]`, set OPENAI_API_KEY in env, then restart Claude Desktop. Claude should show the agent-mcp server in the conversation.

Why does it use short-lived agents?

Each agent is purpose-built for a single task with minimal, focused context. This keeps behavior deterministic, reduces hallucination and the attack surface, and avoids the context bloat of long-running agents. The framework enforces a maximum of 10 active agents and automatic cleanup of idle/finished agents.

Where do I find the admin token?

The admin token is displayed in the server startup logs when the MCP server starts. Workers must be initialized with the worker token (not the admin token).

#multi-agent#orchestration#knowledge-graph#rag#task-management#ai-coding#claude-desktop

Alternatives to Agent-MCP

Compare all alternatives →
Playwright MCP Server
Developer Tools
24k

Microsoft's official browser-automation MCP using Playwright's accessibility tree (no vision model).

Featured
Verified
stdio (local)
No auth
TypeScript
12 tools
Updated 13 days agoRepo
Context7 MCP Server
Developer Tools
28k

Up-to-date, version-specific library documentation injected into your coding agent.

Verified
stdio (local)
API key
TypeScript
2 tools
Updated 17 days agoRepo
Serena
Developer Tools
12k

LSP-powered coding agent toolkit: semantic symbol search, references and structural edits.

Verified
stdio (local)
No auth
Python
11 tools
Updated 15 days agoRepo

Compare Agent-MCP with:

vs Playwright MCP Servervs Context7 MCP Servervs Serenavs Mobile MCP