Verified

Couchbase MCP Server

Couchbase cluster data access for AI models using SQL++ and key-value tools.

Data last scanned today · Reviewed today

634 dangerous tools35 tools34PythonApache-2.0

Overview

Couchbase MCP Server acts as an interface between AI assistants and your Couchbase clusters. It enables agents to execute SQL++ queries, perform key-value operations, and retrieve cluster performance or health metrics. The server is designed to work with both self-managed installations and Couchbase Capella via standard MCP transports like STDIO and HTTP. Built in Python, the server includes administrative controls such as an optional read-only mode to prevent accidental data modifications. Users can manage interaction scopes through fine-grained tool configuration, ensuring AI agents only access the specific database functions required for their current tasks.

Our verdict

With a high-risk security score of 43/100, this tool should be treated with caution in production environments. Crucially, the authentication method and read-only mode implementation have not yet been editorially reviewed, meaning users should rely on manual configuration and rigorous environment isolation. It is best suited for development or internal prototyping by teams comfortable with managing its security posture.

  • Read-only mode available by default
  • Zero direct dependencies reduces supply-chain surface
  • Supports both STDIO and Streamable HTTP transports
  • Active development with frequent recent commits
  • Not maintained by the official vendor
  • Authentication method lacks independent review
  • Read-only mode effectiveness is not yet verified

Tools

ToolDescriptionRisk
get_server_configuration_statusGet the server status and configuration without connecting to the cluster — reports read-only mode, disabled/confirmation-required tools, OAuth settings, and the resolved logging configurationlow
test_cluster_connectionCheck the cluster credentials by connecting to the clusterlow
get_cluster_health_and_servicesGet cluster health status and list of all running services, optionally filtered to specific services via service_typeslow
get_cluster_diagnostics_reportGet the SDK's cached connection diagnostics — whether connections were already broken and for how long, without any active network probinglow
get_buckets_in_clusterGet a list of all the buckets in the clusterlow
get_scopes_in_bucketGet a list of all the scopes in the specified bucketlow
get_collections_in_scopeGet a list of all the collections in a specified scope and bucket. Note that this tool requires the cluster to have Query service.low
get_scopes_and_collections_in_bucketGet a list of all the scopes and collections in the specified bucketlow
get_schema_for_collectionGet the structure for a collectionlow
create_scopeCreate a new scope in a bucket (Couchbase Server 7.6+ and Capella).medium

Compatibility

ClientLocalDockerRemoteRead-only
Claude Desktop
Cursor
VS Code
Windsurf
ChatGPT

Frequently asked questions

What are the system requirements for running the Couchbase MCP Server?

The server requires Python 3.10 or higher. It can be deployed as a Python package from PyPI or via Docker.

Is the server safe to use with AI agents that might inadvertently modify data?

Yes, the server is configured in read-only mode by default, which blocks document mutations and data-modifying SQL++ queries. You must explicitly set the environment variable CB_MCP_READ_ONLY_MODE to false to enable write operations.

Which communication transports does the server support?

The server supports both STDIO and Streamable HTTP (SSE) transport modes.

Can I prevent an AI agent from using specific Couchbase tools?

Yes, you can configure the server to disable individual tools or gate them behind user confirmation.

What Couchbase hosting environments are compatible with this server?

The server connects to both Couchbase Capella-hosted clusters and self-managed Couchbase clusters.

Changelog

Badge

Couchbase MCP Server security score, rated on RepoAI

Maintain this server? Add the live badge to your README.