CockroachDB

You can run a CockroachDB MCP Server to let AI assistants query and interact with your CockroachDB cluster using natural language. It exposes powerful capabilities like schema discovery, CRUD operations, cluster health checks, multi-region queries, and safe transaction handling, all driven through an MCP client.

How to use

Connect to the CockroachDB MCP Server from your MCP client by starting the local server process and providing the required environment details for your CockroachDB cluster. Once connected, you can ask questions in plain language to explore schemas, read data, modify rows within safety limits, and monitor cluster health. Common workflows include asking for available tables, describing a specific table, querying top results, checking cluster status, and performing CRUD operations within configured safety constraints.

Example usage patterns you can try after you have the server running:

• Describe what tables exist in your database
• Inspect the columns of a specific table
• Retrieve a subset of rows from a table with a limit
• Begin a transaction, perform inserts or updates, then commit or rollback
• Check overall cluster health and node status
• Export query results to JSON or CSV for offline analysis

How to install

pip install cockroachdb-mcp

Prepare the environment for your CockroachDB connection by setting the required variables in your shell or a dedicated environment file. You will configure the MCP server to point at your CockroachDB cluster and set the default access mode.

# Example environment variables (fill with your values)
export CRDB_HOST=your-cluster.cockroachlabs.cloud
export CRDB_USER=your-username
export CRDB_PASSWORD=your-password
export CRDB_DATABASE=your-database
export CRDB_CLUSTER=your-cluster-id
export CRDB_READ_ONLY=true

Start the MCP server locally. The server is launched as a standard Python process that the MCP client can connect to using the provided configuration. If you prefer a dedicated HTTP interface or streaming support for clients like Claude, start the appropriate server mode as described in deployment options.

Additional content

Configuration Inspect and safety controls help prevent unintended changes. You can enable read-only mode to block write operations, set a maximum number of rows returned per query, and define a blocklist of disallowed commands. The server supports transaction semantics, so you can begin a transaction, perform reads and writes, and either commit or rollback as needed.

Deployment options include running locally or exposing an HTTP/SSE or streamable HTTP endpoint for clients that require real-time streaming responses. You can also integrate OAuth for SSO with Claude.ai when deploying as a Custom Connector.

Security considerations include ensuring credentials are not logged, enforcing SSL/TLS encryption, and using robust access controls for your cluster. Configure environment variables and runtime options to align with your security requirements.

Example configuration snippet for the MCP server in a client workflow demonstrates the required runtime command and environment settings. Use this as a template to connect your Claude Desktop or other MCP clients.

Available tools

connect

Establish a connection to the CockroachDB cluster from the MCP client.

disconnect

Close an active MCP connection to the cluster.

cluster_status

Retrieve current health and status information for the CockroachDB cluster.

list_databases

List all databases available in the connected cluster.

list_tables

List tables and views in the selected database.

describe_table

Get column information for a specific table.

execute_query

Run SELECT queries and return results.

validate_query

Check query safety prior to execution.

read_rows

Read rows by key or filter.

insert_row

Insert a new row into a table.

update_row

Update existing rows in a table.

delete_row

Delete rows by key.

begin_transaction

Start a new transaction.

commit_transaction

Commit the current transaction.

rollback_transaction

Rollback the current transaction.

export_to_json

Export query results to JSON.

export_to_csv

Export query results to CSV.

save_knowledge

Persist learned information about the cluster.

get_all_knowledge

Retrieve all saved knowledge.