Couchbase Guru MCP Server

An MCP server that lets LLMs search the Couchbase documentation from your MCP client. It exposes a single tool, ask_couchbase_docs, which forwards your question to a hosted retrieval-augmented (RAG) documentation agent and returns an answer with source links.

No Couchbase cluster or credentials required. The server talks only to the documentation agent backend, not to your data.

Tool

Ask complete, self-contained questions — the backend has no conversation history, so include the product, version, and language where relevant (e.g. "How do I create a primary index with the Python SDK in Couchbase Server 7.6?").

Prerequisites

• Python 3.10 or higher.
• uv installed to run the server.
• An MCP client such as Claude Desktop, Cursor, or VS Code.

Configuration

The server can be run from the prebuilt PyPI package or from source with uv. It works with zero configuration — the public documentation agent is used by default.

Running from PyPI

{
  "mcpServers": {
    "couchbase-guru": {
      "command": "uvx",
      "args": ["couchbase-guru"]
    }
  }
}

If you already have other MCP servers configured, add this entry to the existing mcpServers object.

Running from Source

Clone the repository:

git clone https://github.com/Couchbase-Ecosystem/couchbase-guru.git

Then point your MCP client at it:

{
  "mcpServers": {
    "couchbase-guru": {
      "command": "uv",
      "args": [
        "--directory",
        "path/to/cloned/repo/couchbase-guru/",
        "run",
        "src/mcp_server.py"
      ]
    }
  }
}

path/to/cloned/repo/couchbase-guru/ should be the path to the cloned repository on your machine. Don't forget the trailing slash.

Options

All options are optional and can be set via CLI argument or environment variable:

Check the installed version with:

uvx couchbase-guru --version

Self-hosting the documentation agent

By default the server uses a shared, public documentation agent, so most users need no setup. If you run your own agent backend, point the server at it:

uvx couchbase-guru --agent-base-url https://your-agent.example.com

Rate limiting & privacy

The public agent applies fair-use rate limits. To support this, the server sends a pseudonymous device identifier to the backend (in the User-Agent header):

• stdio: a random id generated once and stored in a per-user file on your machine.
• HTTP: a salted, one-way hash of the connecting IP — the raw address is never sent.

No question content or personal data is persisted by the MCP server itself. If you prefer not to share a rate-limit signal, self-host the agent (see above).

Client-specific configuration

Streamable HTTP transport mode

The server can run in Streamable HTTP mode so multiple clients can connect to one instance. Check that your MCP client supports this transport first.

uvx couchbase-guru --transport=http --port=8000

The server will be available at http://localhost:8000/mcp:

{
  "mcpServers": {
    "couchbase-guru-http": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

This mode does not include authorization support.

Docker

Build the image:

docker build -t couchbase-guru .

Run it (stdio by default; no credentials needed):

{
  "mcpServers": {
    "couchbase-guru-docker": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "couchbase-guru"]
    }
  }
}

For HTTP transport, publish the port and set the transport:

docker run --rm -i \
  -e CB_MCP_TRANSPORT=http \
  -e CB_MCP_HOST=0.0.0.0 \
  -e CB_MCP_PORT=8000 \
  -p 8000:8000 \
  couchbase-guru

Risks associated with LLMs

• The use of large language models and similar technology involves risks, including the potential for inaccurate or harmful outputs.
• Couchbase does not review or evaluate the quality or accuracy of such outputs, and such outputs may not reflect Couchbase's views.
• You are solely responsible for determining whether to use large language models and related technology, and for complying with any applicable license terms, terms of use, and your organization's policies.

Troubleshooting

• Confirm that uv/uvx is installed and on your PATH. You may need to provide an absolute path to uv/uvx in the command field.
• If a search times out, the documentation backend may be busy — retry in a moment.
• To rule out the public backend, run against your own agent with --agent-base-url.
• If running from source after updating the repo, run uv sync to refresh dependencies.
• Check your MCP client's logs (locations above) for errors.

Testing

Unit tests run offline (the backend is mocked):

uv sync --extra dev
uv run pytest tests/

Integration tests exercise the tool end-to-end against a live agent backend and are opt-in:

CB_MCP_RUN_INTEGRATION=1 uv run pytest tests/test_docs_tools.py

By default they use the public agent; set CB_AGENT_BASE_URL to target a different backend.

👩‍💻 Contributing

Contributions are welcome! To report a bug, request a feature, or contribute improvements, open a GitHub issue.

See CONTRIBUTING.md for developer setup (environment with uv, linting/formatting with Ruff, pre-commit hooks, and project structure).

# Clone and set up
git clone https://github.com/Couchbase-Ecosystem/couchbase-guru.git
cd couchbase-guru

# Install with development dependencies
uv sync --extra dev