coze

The COZE MCP Server provides a set of tools to manage and interact with COZE bots and workspaces via the Model Context Protocol. Available tools include listing workspaces and bots, creating and updating bots, retrieving your own user info, publishing bots to an API channel, and interacting with bots and workflows through chat. You can also query available voices. To use the server, deploy it with one of the supported runtimes (uvx, Python, or Docker) and point your MCP client to the configured mcpServer entry. Once connected, use the provided tool commands to manage resources, such as listing workspaces with list_workspaces, or creating a bot with create_bot, then chat with that bot using chat_with_bot to obtain responses in context-aware sessions.

Prerequisites:

• Python 3.10+ or an installation of uv (for uvx) or Docker if you prefer the container path.
• PIP if installing via Python package.
• Internet access to fetch dependencies.

Install via pip (recommended if you are not using uv/containers):

pip install coze-mcp-server

Run as a module:

python -m coze_mcp_server

Install with uv (recommended):

# uvx will run the server directly
# No extra install required beyond uv/uvx tooling

Run with Docker:

# Ensure Docker is running and you have network access
docker run -i --rm ghcr.io/coze-dev/coze-mcp-server

Configuration examples (as shown in README):

• uvx (via Claude/uv-based workflow):

"mcpServers": {
  "coze-mcp-server": {
    "command": "uvx",
    "args": ["coze-mcp-server"]
  }
}

• Docker:

"mcpServers": {
  "coze-mcp-server": {
    "command": "docker",
    "args": ["run", "-i", "--rm", "ghcr.io/coze-dev/coze-mcp-server"]
  }
}

• Python module (pip install):

"mcpServers": {
  "coze-mcp-server": {
    "command": "python",
    "args": ["-m", "coze_mcp_server"]
  }
}

Tips and common issues:

• Ensure the environment where the MCP server runs has network access to the COZE APIs if your bots connect to external services.
• When using Docker, adjust resource limits (CPU/m memory) if you encounter timeouts or slow responses.
• If using uvx, ensure the coze-mcp-server binary is resolvable in your PATH; the tool will typically download/run the package directly.
• For authentication or API access, consider setting related environment variables as required by your deployment (e.g., API keys or endpoint URLs).
• The available tools can be invoked in your MCP client by name (list_workspaces, list_bots, create_bot, get_me, retrieve_bot, update_bot, publish_bot, chat_with_bot, chat_with_workflow, list_voices). Use these as documented by your MCP client to perform the corresponding actions.