Connect an MCP client

Lainstack runs an MCP server, so an AI client can read and build automations, look at runs and handle escalations in your workspaces, with your permissions. This guide is for developers and for anyone who works in an AI coding or chat tool.

You sign in once through your browser. The client never sees your password, and you choose which workspaces it can reach.

1. Get your MCP URL

In Lainstack, open Personal, then Connections, and click Connect a client. The dialog shows the MCP URL for your account, which ends in /api/mcp, and the exact command for each client. Copy the URL from there.

2. Add Lainstack to your client

Use the instructions for your client, replacing <MCP URL> with the URL from step 1.

Claude Code

claude mcp add --transport http lainstack <MCP URL>

Add --scope user to use it in every project. Then run /mcp in Claude Code and choose to authenticate.

Codex

codex mcp add lainstack --url <MCP URL>
codex mcp login lainstack

Cursor

Use Add to Cursor in the Lainstack dialog, or add the URL to ~/.cursor/mcp.json. Then open Cursor’s MCP settings and log in.

Other clients

Any client that supports MCP over Streamable HTTP with OAuth can connect. Give it the URL; it finds the sign-in details on its own and registers itself.

3. Approve the connection

You sign in once in your browser. The client never sees your password, only a token limited to the workspaces you chose.

Your client opens a Lainstack page in your browser. If you are not signed in, sign in first; the connection is made for the organization you are signed in to.

The page shows the client’s name, the Permissions requested in plain language, and Workspaces this connection can access. Tick the workspaces the client may use, at least one, and click Approve and connect.

The client can now call Lainstack. It sees only the tools its permissions allow, and only in the workspaces you ticked. Your own role in each workspace still applies: the client cannot do anything you could not do yourself.

What a client can do

The tools are grouped by area: automations, runs and their evidence, escalations, operations, reference tables, workspaces, and the product guides. Each tool is marked by what it changes: read only, draft edits, publishing and activating, run control, consequential writes, or destructive. Clients can use this to ask you before acting.

A client can start a live run and read runs, their steps and their values. Some run actions stay inside Lainstack, for people in the app and the in-app Agent: running a draft, resuming a failed run, and failure groups, including proving a fix. A client does not see these tools. The public API does not offer them either.

Stay connected or disconnect

Your client signs in again in the background as needed. If it stays unused for 30 days, you approve it again.

To see or cut off access, open Personal, then Connections. Each client shows its scopes, workspaces, and when it was connected and last used. Revoke stops it at once. To give a client more workspaces, approve the connection again.

API keys do not work with the MCP server. Sign in as shown above.