Skip to main content
Every connector has its own MCP endpoint (streamable HTTP). Point an IDE or an agent at it and you can browse that server’s databases, run bounded SQL and explore its shares, without writing curl.
The connector is identified by the URL on every request, so a session can never drift to another server. Add one entry per connector.

Set it up in Cursor

Add this to ~/.cursor/mcp.json, or to .cursor/mcp.json in a project:
To find <connector_id>:
The API key in mcp.json is an organization-wide key. Protect the file like a credential and never commit it to version control. Keys limited to one connector are planned.

Tools

Suggested order for a database: connector_status, list_connections, list_tables, list_columns or describe_table, then query. Suggested order for files: connector_status, list_filesystems, fs_du on a candidate folder, fs_ls, then fs_search if you need to look inside. Try a prompt such as: “Using the timbal-connector-erp tools, list the tables in erp and describe the ones about customers.”

Behaviour

  • The tools do exactly what the equivalent HTTP requests do: same limits, same capability checks, same audit entry for query (marked via=mcp).
  • The connection’s query setting on the server still decides what query may do.
  • Every tool except connector_status needs the connector online; connector_status still answers when it is offline. A connector that is offline, or lacks a capability, returns a tool error with the same message as the API. Nothing here can reach a server the connection can’t.
  • It does not start jobs or syncs, copy files into a knowledge base, or push configuration. For those, use the API.

Next steps

Sync your data

Start a job or schedule a sync.

MCP integration

Connect Timbal’s own tools to your IDE.