Connect your coding agent

Install and authenticate the Laimonade MCP server in Claude, Claude Code, Cursor, Codex or any Streamable HTTP client, so your coding agent can read the sprint and hand work back.

Last reviewed · maintained by Founder

What connecting gives you

A connected coding agent can read your Ready column, load an item with its full acceptance criteria and regression notes, file the bugs it finds along the way, and hand finished work back for review. Nobody copies tickets into a prompt.

The connection is one endpoint, spoken over the Model Context Protocol:

https://api.laimonade.com/mcp

Which clients work?

Any MCP client that supports Streamable HTTP. The four we test are Claude (web, desktop and mobile), Claude Code, Cursor and Codex.

How do I authenticate?

Two credentials are accepted on the same endpoint, and which one you use decides whose name the work is recorded under:

  • Sign in (OAuth). Recommended for anything a person is driving. The session is bound to you, so milestones your agent creates are owned by you and a blocker it reports says whose agent hit it.
  • Project API key. For headless use — CI, cron, scripts. It identifies a project, not a person, so work lands on the project owner. Find it under MCP in the app sidebar.

Never commit an API key. If one leaks, regenerate it from the same page — every configured client then has to be updated, which is the reason to prefer signing in.

Claude (web, desktop, mobile)

Laimonade is an official Claude connector, so there is nothing to paste.

  1. In Claude, open Settings → Connectors and find Laimonade.
  2. Click Connect. A Laimonade page opens where you sign in and choose which project or projects to grant.
  3. Approve. The connector now works across Claude web, desktop and mobile.

Claude Code

Add the server, then authenticate from inside the session:

claude mcp add --transport http laimonade https://api.laimonade.com/mcp

Run /mcp inside Claude Code, choose laimonade → Authenticate, and your browser opens the Laimonade consent page. Run /mcp again to confirm it shows as connected.

The server name is a local label — call it whatever you like; it is what you will see in /mcp output.

To share the configuration with your team, commit a .mcp.json in the repo root instead:

{
  "mcpServers": {
    "laimonade": {
      "type": "http",
      "url": "https://api.laimonade.com/mcp"
    }
  }
}

For headless use, add --header "Authorization: Bearer YOUR_PROJECT_API_KEY" to the claude mcp add command. That skips the sign-in flow and gives up per-person attribution.

Cursor

Add this to .cursor/mcp.json in your repo, or use Cursor Settings → MCP for a global configuration:

{
  "mcpServers": {
    "laimonade": {
      "url": "https://api.laimonade.com/mcp"
    }
  }
}

Restart Cursor or reload MCP servers from settings. Cursor prompts you to sign in on first connect. For headless use, add a headers object carrying "Authorization": "Bearer YOUR_PROJECT_API_KEY".

Codex and other MCP clients

Any client with Streamable HTTP transport can connect directly:

  • POST initialize to the MCP URL, with no session header on the first call.
  • Send Authorization: Bearer <credential> on every request.
  • Read the Mcp-Session-Id response header and send it on follow-up calls.

https://api.laimonade.com/mcp/health needs no authentication and reports the running server version, which is the quickest way to tell a network problem from an auth problem.

Can one connection cover several projects?

Yes. A single sign-in can be granted more than one project. When it is, every tool call names its target project, and every reply is stamped with the project that answered — so nothing can be created in the wrong backlog without it being visible.

Ask your agent to call list_projects to see what the connection was granted. You can change the granted set at any time from the MCP page in the app, with no reconnect.

How do I know it worked?

Ask your agent:

What's ready this sprint?

You should get a numbered list of the Ready column. If the list is empty, that is a real answer rather than a failure — it means nothing is Ready on this week's sprint. Check the board.

In the app, the MCP page shows when the credential was last used, which is the other half of the confirmation.

The loop, once you are connected

  1. Ask what is ready. A numbered menu of the Ready column comes back. There are narrower versions for stories, bugs and improvements, and one grouped by epic.
  2. Load the items you want. "Load 1 and 3" pulls them in full — description, acceptance criteria, estimate, parent epic.
  3. Get the context before changing anything. Ask for the regression notes covering the paths about to be edited, and for the project's own rules — your CLAUDE.md, AGENTS.md and Cursor rules.
  4. Hand it back. "Submit 1 for review" moves the item to In Review. Submitting asserts the acceptance criteria are met, which is why loading the item first matters.

Menu numbers are session state: they always refer to the most recent list. If a reference looks wrong, ask for the menu again rather than guessing.

What an agent cannot do

It cannot mark work done, cannot add anything to the unattended-implementation queue including its own items, cannot reach another project, and cannot read your Claude memory, chat history or uploaded files. The privacy policy states what is stored; what Laimonade is explains why those limits are deliberate.

Where to go next

The MCP tool reference lists every tool with its permission. If a connection misbehaves, troubleshooting covers the failures people actually hit.