Connect your agent

QueryLane includes an MCP server. An AI agent that supports MCP, such as Claude Code or Cursor, can use it to read and build jobs, run queries and runs, and inspect results in your QueryLane profile. The agent uses its own model and account. The built-in assistant is not needed for this.

Connect a supported client

QueryLane can set up these clients for you: Claude Code, Claude Desktop, Codex, Cursor, VS Code, Z Code, OpenCode, Hermes Agent and OpenClaw.

1. Install the client on your Mac.
2. In QueryLane, open Settings > CLI and MCP.
3. Find the client under QueryLane MCP and click Connect. Clients that QueryLane does not find on your Mac are listed under Other supported clients. After you install one, click Refresh.
4. Restart the client or start a new session in it. QueryLane reminds you of this after it saves the configuration.
5. Check in the client that the QueryLane tools are available. Their names start with querylane_.

Connect installs the QueryLane CLI at ~/.local/bin/querylane if it is missing, and adds a querylane server entry to the configuration of the client. It keeps the other entries in that file. For VS Code, the entry goes to the default VS Code profile.

The status next to each client shows Connected, Not connected, Update required or CLI required. For a client with Update required, click Update connection. If QueryLane warns that the CLI points to an older app location, click Update CLI.

QueryLane has no button to disconnect a client. To disconnect, remove the querylane entry from the client configuration.

Configure another client

For a client that is not in the list but runs local stdio MCP servers:

1. Open Settings > CLI and MCP.
2. If the QueryLane CLI is not installed, click Install CLI.
3. Under Other MCP client, click Show settings.
4. Copy the Command and Arguments, or the Configuration example, into the MCP settings of your client.
5. Restart the client or start a new session.

The example has this shape:

{
  "mcpServers": {
    "querylane": {
      "command": "/Users/you/.local/bin/querylane",
      "args": ["mcp", "serve", "--profile", "main"]
    }
  }
}

The client starts the server with the stdio transport. --profile selects the QueryLane profile the agent works in. Some clients use another top-level key instead of mcpServers, such as servers, mcp_servers, mcp or context_servers.

What the agent can do

The agent gets every tool of the QueryLane MCP server. With them it can:

• read jobs, connections, database schemas, runs, results and plugins;
• create jobs and connections, propose a change to a job with a diff, apply it and revert it;
• test steps, start and cancel runs, and check results;
• run SQL or MongoDB queries through a saved connection;
• export data to CSV, JSON, JSONL or XLSX, and render charts;
• call tools of installed plugins;
• open a job, step, connection or settings section in the QueryLane window.

The MCP tools reference lists every tool with what it reads, changes or runs.

Approvals

QueryLane does not ask for approval when an agent calls a tool over MCP. The approval buttons of the built-in assistant do not apply here. Your client decides whether to ask you, based on its own permission settings.

Set these permissions before you give the agent a task:

• allow the read tools if you want the agent to explore on its own;
• keep confirmation for the tools that change or execute something, such as querylane_db_query_execute, querylane_change_apply or querylane_run_start.

A query runs with the database user of the connection, so the agent can do whatever that user is allowed to do. For a database the agent should only read, connect with a read-only database user.

The agent receives real data from queries and results. Data for AI applies only to the built-in assistant, see Data privacy.