MCP

Hand findings from Alloy to your coding agent and stay in your editor from the bad run to the commit. Connect Claude Code, Codex or Cursor to Alloy once, over MCP.

Setup: you connect your coding agent with one command.

What to use it for

• Pull the mission, the agent's finding and the data behind each plot into the session where you change the code.
• Compare a bad run with the last good run on the previous build.
• Write the fix back to Alloy as a report, linked to the missions.
• Set up a report automation that checks every new run.

Example

Fictional example.

An engineer opens Claude Code in the flight-controller repository and asks: "Pull Alloy's report for HX-12 flight F-0412. Get the altitude data around the oscillation and from the last good flight. Then find what changed in our altitude controller." Their coding agent reads the finding and the data from Alloy, then searches the code.

How to use it

Ask in plain language. Your coding agent picks the Alloy tools it needs.

Set up

For most organisations the server URL is below. If you sign in to Alloy at a different address, use that address followed by /mcp.

https://aus.usealloy.ai/mcp

• Claude Code: run claude mcp add alloy --transport http https://aus.usealloy.ai/mcp, then run /mcp and sign in.
• Codex: run codex mcp add alloy --url https://aus.usealloy.ai/mcp, then codex mcp login alloy.
• Cursor: add the server URL to .cursor/mcp.json and restart Cursor.
• Claude Desktop: open Settings → Connectors, click Add custom connector and paste the server URL.
• Other clients, such as Windsurf: use the server URL with Streamable HTTP and OAuth sign-in in the browser.

Copy this into your AGENTS.md or CLAUDE.md

Paste the Alloy rules snippet into the instructions file in your robot software repository. It tells your coding agent when to use Alloy and how.

## Robot data: use Alloy

Our robot runs live in Alloy (MCP server "alloy"). Use it when a task involves
a run, flight, mission, robot log, field issue or test result.

When to use Alloy:
- Before you change control, estimation, perception or planning code because of
  a field or test issue, pull the mission and the agent's report from Alloy.
- Use Alloy for robot run data. Do not use product analytics or guess from code.

How to use it:
- Start from the mission. Name the robot, the run ID or the time window.
- Get the data points for the window you need. Do not download whole recordings
  unless the task needs raw data.
- Compare against the last good run on the previous build.
- Cite mission IDs and timestamps for every claim about robot behaviour.
- When the fix is done, create an Alloy report: the cause, the evidence, the
  fix and the commit link. Link the missions.
- If the issue can come back, propose a report automation that checks new runs.
  Ask me before you create it.

Stop and ask me when:
- An Alloy sign-in, API key or upload is needed.
- You need to delete, rename or archive anything in Alloy.
- The Alloy tools you expected are missing. Our organisation may not have that product.
- An Alloy tool returns "Could not resolve organization". I belong to more than one
  organisation and must set the organisation URL.

Hand-off to the agent

If your organisation has the Agent Platform but not Mesh Storage, your coding agent also sees a hand-off tool. Set the hand-off tool to ask first in your coding agent.

Related