Lightweight Python adapter that makes ACP agents discoverable and invocable from any MCP-compatible client (e.g., Claude Desktop) by exposing each agent as an MCP resource and a run_agent tool.
https://github.com/i-am-bee/acp-mcpStop choosing between AI ecosystems. This lightweight adapter connects your ACP agents directly to MCP-compatible applications like Claude Desktop, giving you the best of both worlds without rebuilding anything.
You've built agents using the Agent Communication Protocol (ACP) for agent-to-agent workflows. Now you want to use them in Claude Desktop or other MCP applications. Your options? Rewrite everything for MCP, maintain parallel implementations, or manually copy-paste outputs between systems.
There's a better way.
The ACP-MCP adapter creates a seamless bridge between these ecosystems. Point it at your ACP server, and your agents instantly become available as MCP resources and tools.
uvx acp-mcp http://localhost:8000
That's it. Your ACP agents now appear in Claude Desktop's resource picker, and you can invoke them with natural language commands like:
use "data-processor" agent with the customer CSV file
Faster Agent Testing: Instead of switching between terminals and manually formatting inputs, test your ACP agents directly in Claude Desktop's conversational interface.
Unified Workflows: Use your specialized ACP agents alongside Claude's reasoning capabilities without context switching or manual data transfer.
Zero Migration Overhead: Keep your existing ACP infrastructure while gaining access to the growing MCP ecosystem.
Development Teams with existing ACP agents who want Claude Desktop integration without rewrites. Connect in minutes, not weeks.
Multi-Agent Orchestration where you need ACP agents to participate in broader MCP-driven workflows. The adapter handles protocol translation automatically.
Rapid Prototyping when you want to quickly test how your ACP agents perform in conversational contexts before building dedicated UIs.
This adapter prioritizes simplicity and speed over feature completeness:
If you need full-featured agent-to-agent communication, stick with pure ACP. If you need quick access to your agents in MCP applications, this adapter delivers exactly what you need.
Add this to your Claude Desktop config:
{
"mcpServers": {
"acp-local": {
"command": "uvx",
"args": ["acp-mcp", "http://localhost:8000"]
}
}
}
Restart Claude Desktop. Your ACP agents appear as attachable resources and invokable tools immediately.
The adapter auto-discovers your registered ACP agents and exposes them with the URI pattern acp://agents/{agent_name}. No manual configuration, no agent registration - just connect and use.
Ready to bridge your AI workflows? Clone the repo, point it at your ACP server, and start using your agents in Claude Desktop today. Built by the BeeAI team as part of the Linux Foundation AI & Data program.