MCP 服务器

MCP Server

A minimal reference implementation of the Model Context Protocol (MCP) with a few mock tools and an optional Gradio playground.

Usage

Run the server locally:

python server.py

Create a dev.env file and add your OpenAI API key:

OPENAI_API_KEY=your-openai-key

Inside Google Colab or other notebooks use the helper:

from colab_adapter import launch_in_colab
launch_in_colab()

The API is served on port 8000 by default and the Gradio UI will try to use GRADIO_SERVER_PORT or the first free port starting at 7860.

Run in Google Colab

Open MCP_colab.ipynb in Colab or click the badge above and run the cells.

If starting from a blank notebook, run these commands to set up and launch the server:

!git clone https://github.com/zack-dev-cm/mcp_server.git
%cd /content/mcp_server
!pip install fastapi uvicorn[standard] gradio==4.* pydantic python-dotenv \
            httpx openai pydantic-settings
from colab_adapter import launch_in_colab
launch_in_colab()

You can now query the API from another cell:

import requests, time
time.sleep(2)
print(requests.get("http://localhost:8000/v1/resources").json())

The server output shows a public URL for the Gradio interface so you can try the demo visually.

LLM/VLM Plugin Example

Plugins can extend the server with new tools. The included openai_chat and openai_vision plugins show how to call OpenAI models. Set OPENAI_API_KEY in your environment and start the server.

If running locally, install the required packages first:

pip install openai httpx pydantic-settings
python server.py

The Dockerfile installs openai automatically. Invoke the openai.chat or openai.vision tools via the API or Gradio UI.

Streamable HTTP Endpoint

ChatGPT connectors can talk directly to the server using the unified /mcp route. Send JSON‑RPC requests with POST /mcp to receive standard JSON responses. When your client supports Server‑Sent Events you may instead GET /mcp and keep the connection open to stream updates.

Configure your ChatGPT connector to point at your server’s base URL and use the /mcp endpoint for both non‑streaming and streaming interactions.

User Data API

Authenticated sessions can store and manage user‑specific JSON payloads. First request a session ID then pass it as a bearer token:

# create a session and grab the token
TOKEN=$(curl -s -X POST http://localhost:8000/v1/initialize \
  -H 'Content-Type: application/json' \
  -d '{"id":1,"jsonrpc":"2.0","method":"initialize","params":{}}' \
  | jq -r '.result.sessionId')

# store data
curl -X POST http://localhost:8000/api/user/data \
  -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"foo":"bar"}'

# fetch it back
curl -H "Authorization: Bearer $TOKEN" http://localhost:8000/api/user/data

# remove it
curl -X DELETE -H "Authorization: Bearer $TOKEN" \
  http://localhost:8000/api/user/data

Deploying to Cloud Run

You can deploy the server on Google Cloud Run using the provided Dockerfile:

# build and push the container
gcloud builds submit --tag gcr.io/PROJECT_ID/mcp-server

# deploy the image to Cloud Run
gcloud run deploy mcp-server \
  --image gcr.io/PROJECT_ID/mcp-server \
  --region REGION \
  --allow-unauthenticated

Cloud Run sets the PORT environment variable automatically, which the server uses to expose both the API and Gradio UI on the same endpoint.

Running Tests

The test suite expects both ELEVENLABS_MCP_SECRET and OPENAI_API_KEY to be set. Copy dev.env and source it before executing pytest:

cp dev.env .env
set -a && source .env && set +a
pytest