Hand off a task to Cursor

The same loop as the Claude Code guide (pull, claim, work, report, hand back), driven from Cursor. This guide covers what’s different; the loop itself is client-agnostic, which is the point of MCP.

1. Connect

Add Masaro to Cursor’s MCP configuration:

json

{
  "mcpServers": {
    "masaro": { "url": "https://masaro.ai/api/mcp/mcp" }
  }
}

Cursor picks up the config and starts the OAuth flow on first use: browser → sign in → consent screen for your workspace → approve. Check Cursor’s MCP panel: masaro listed, tools discovered.

Cursor’s MCP settings UI moves between releases. If the panel looks different from a screenshot you’ve seen, the JSON above is the stable truth.

2. Hand off from the agent panel

In Cursor’s agent chat, with your project open:

text

Get the next task from Masaro, claim it, and work it in this repo.
Report progress after each meaningful step. If anything is ambiguous,
report a blocker instead of guessing.

Cursor’s agent calls the same tools Claude Code does (get_next_task, claim_task, update_progress) because it’s the same server. Your board shows the claim and the progress feed identically.

3. Mind the approval settings

Cursor asks for per-tool approval by default. Approve the Masaro tools when prompted, or pre-allow them in Cursor’s tool settings if you’re comfortable; the write-side tools are guarded on Masaro’s end anyway (authoring lands in Triage, completion goes to In Review, nothing self-approves into Done), under the default supervised policy (a workspace can widen it in settings).

4. Review as usual

The hand-back is identical: summary plus per-criterion mapping, task in In Review, your grade decides. The board doesn’t know or care which client did the work. That’s what “any MCP client” means in practice.

Also works: any MCP client with remote-server (streamable HTTP + OAuth) support. Same URL, same flow, same board.