Skip to main content

Use Loha with Claude Code & Cursor

Loha's MCP server lets your AI coding agent talk to Loha directly. Instead of switching to the dashboard when a build goes red, you ask the agent, and it reads the failing logs and suggests a fix in the file you already have open.

This guide gets you connected in three steps.

1. Create a personal access token​

In the dashboard, open your workspace's Settings → API Keys and create a personal access token (the dashboard calls it an API key). Members, admins and owners can create one (guests cannot), where member self-service is enabled for the workspace; starting builds (triggers:run) needs the member role or higher. Scope it to what you want the agent to do:

  • Read-only (builds:read, repos:read, pipelines:read): inspect builds, read logs, validate configs. The safe default.
  • Add triggers:run only if you want the agent to be able to start builds.
Default to read-only

An agent runs with your token's permissions, so a read-only token keeps it strictly "look, don't touch." It's the simplest way to make the agent's access narrower than your own.

Copy the token (loha_…); you'll paste it next.

2. Add the MCP server​

Claude Code:

claude mcp add --transport http loha https://mcp.loha.longtumlab.com/ \
  --header "Authorization: Bearer loha_xxxxxxxx"

Cursor / Claude Desktop: add to your MCP config (.mcp.json or the app's config):

{
  "mcpServers": {
    "loha": {
      "type": "http",
      "url": "https://mcp.loha.longtumlab.com/",
      "headers": { "Authorization": "Bearer loha_xxxxxxxx" }
    }
  }
}

Restart the agent if needed, then confirm it's connected: ask it "what Loha workspace am I in?" (it calls whoami).

3. Try it​

A few things to ask your agent:

  • "Are my builds green?" → it lists repos and recent builds and summarizes.
  • "Why did the last build on main fail?" → it finds the failed run, opens the failing step, reads the redacted logs, and proposes a fix.
  • "Is this .loha.yml valid?" → it validates against Loha's real schema before you commit.
  • "Run the pipeline for this repo on main." → it shows you the intended run and waits for your go-ahead (guarded; see below).
Starting builds is guarded

trigger_pipeline dry-runs by default: the agent tells you what it would run and dispatches nothing until it confirms. And with a read-only token, it can't start builds at all.

Build data and your model

Logs and build metadata the agent reads enter its model's context. Build logs are secret-redacted by Loha first, but see Data sovereignty before using it on sensitive output.

See also​