A Model Context Protocol (MCP) server that exposes OpenAI agents through the MCP protocol.
This server exposes both individual agents and a multi-agent orchestrator using the OpenAI Agents SDK:
- Web Search Agent: A specialized agent for searching the web for real-time information
- File Search Agent: A specialized agent for searching and analyzing files in OpenAI's vector store
- Computer Action Agent: A specialized agent for performing actions on your computer safely
- Orchestrator Agent: A powerful agent that can coordinate between the specialized agents, choosing the right one(s) for each task
Each agent is accessed through the MCP protocol, making them available to any MCP client, including the Claude desktop app.
- Python 3.11 or higher
- uv package manager (recommended)
- OpenAI API key
To install openai-agents-mcp-server for Claude Desktop automatically via Smithery:
npx -y @smithery/cli install @lroolle/openai-agents-mcp-server --client claude
"mcpServers": {
"openai-agents-mcp-server": {
"command": "uvx",
"args": ["openai-agents-mcp-server"],
"env": {
"OPENAI_API_KEY": "your-api-key-here"
}
}
}
- WebSearchTool: No required parameters, but can accept optional location context
- FileSearchTool: Requires vector_store_ids (IDs from your OpenAI vector stores)
- ComputerTool: Requires an AsyncComputer implementation (currently simulated)
You can customize this server by:
- Implementing a full AsyncComputer interface to enable real computer interactions
- Adding additional specialized agents for other OpenAI tools
- Enhancing the orchestrator agent to handle more complex workflows
You can configure the server using environment variables:
OPENAI_API_KEY
: Your OpenAI API key (required)MCP_TRANSPORT
: Transport protocol to use (default: "stdio", can be "sse")
# Clone the repository
git clone https://github.com/lroolle/openai-agents-mcp-server.git
cd openai-agents-mcp-server
# Create a virtual environment
uv venv
source .venv/bin/activate # On Windows: .venv\Scripts\activate
# Install dependencies
uv sync --dev
You can test the server using the MCP Inspector:
# In one terminal, run the server with SSE transport
export OPENAI_API_KEY=your-api-key
export MCP_TRANSPORT=sse
uv run mcp dev src/agents_mcp_server/server.py
Then open a web browser and navigate to http://localhost:5173.
MIT