OpenReview Python MCP Server

MCP server that helps LLMs write correct openreview-py code. Two knowledge layers: live introspection of the installed library (method signatures, docstrings, class structures) and static knowledge (best practices, code examples, workflow guides).

Tools

Usage

Option 1: Local Docker (stdio)

Build and use directly with Claude Code:

docker build -t openreview-mcp .

Add to .mcp.json:

{
  "mcpServers": {
    "openreview": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "openreview-mcp"]
    }
  }
}

Restart Claude Code (/exit and relaunch). The 5 tools will be available immediately.

Option 2: Local Docker service (SSE)

Run as a persistent local service:

docker run -d --name openreview-mcp -p 8080:8080 openreview-mcp --transport sse

Add to .mcp.json:

{
  "mcpServers": {
    "openreview": {
      "url": "http://localhost:8080/sse"
    }
  }
}

Option 3: Remote server (public)

Deploy to any host that can run Docker (Cloud Run, Fly.io, a VM, etc.):

docker run -d -p 8080:8080 openreview-mcp --transport streamable-http

Clients connect via .mcp.json:

{
  "mcpServers": {
    "openreview": {
      "url": "https://your-server.example.com/mcp"
    }
  }
}

No API token is needed — this server provides knowledge tools only (read-only, no live API access).

CLI options

openreview-mcp [--transport stdio|sse|streamable-http] [--port 8080] [--host 0.0.0.0]

Knowledge files (llm.txt, examples.md) are bundled inside the image — no bind-mount required. To override with a live openreview-py checkout, add -e OPENREVIEW_KNOWLEDGE_PATH=/knowledge -v /path/to/openreview-py:/knowledge to the docker run args.

Reusable Registration

Other FastMCP servers can mount the knowledge tools without running the full server:

from openreview_mcp.registration import register_knowledge_tools

register_knowledge_tools(mcp)  # mounts 5 knowledge tools onto your FastMCP instance

Keeping Up with openreview-py

The introspection layer (search_api, get_method_signature) reads directly from the installed openreview-py library at startup. When openreview-py adds or changes methods, rebuild the Docker image to pick them up:

docker build --no-cache -t openreview-mcp .

The --no-cache flag ensures pip fetches the latest openreview-py from GitHub instead of using a cached layer.

For the static knowledge layer (get_best_practices, get_code_example, get_workflow_guide), update the bundled files in openreview_mcp/knowledge_files/ and rebuild.

Alternatively, to use a live checkout of openreview-py without rebuilding, mount it at runtime:

docker run --rm -i \
  -e OPENREVIEW_KNOWLEDGE_PATH=/knowledge \
  -v /path/to/openreview-py:/knowledge \
  openreview-mcp

This overrides both the introspected library and the knowledge files path.

Development

git clone https://github.com/openreview/openreview-mcp.git
cd openreview-mcp
python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
.venv/bin/pytest tests/ -v