# Connect your coding agent

> Give your coding agent what Stableship found, so it can fix it and ask Stableship to check the fix.

## What connecting does

Your coding agent connects to the Stableship MCP server. Once it is connected, it can:

- Read what Stableship found on your app, and the fix brief for each problem: what is wrong, the rule it breaks, where to look and what must be true once it is fixed.
- Work through the fix plan in order and ask Stableship to verify each fix on its branch.
- Read your rules before it changes sign-in, data access, payments or migrations, and propose a rule change instead of quietly changing who can see what.
- Start a check before it pushes a risky change, and wait for the verdict.

Stableship never writes to your repository. Your agent writes and commits every fix; Stableship checks it.

## Pick your tool

| Tool | Kind | Signs in with | Reads instructions from | Tested end to end |
|---|---|---|---|---|
| [Claude Code](/docs/agents/claude-code) | Coding agent | Plugin, then browser sign-in | `CLAUDE.md`, `AGENTS.md` | Yes |
| [Cursor](/docs/agents/cursor) | Coding agent | Browser sign-in, or a token | `AGENTS.md`, `.cursor/rules` | Not yet |
| [Codex](/docs/agents/codex) | Coding agent | Browser sign-in, or a token | `AGENTS.md` | Not yet |
| [Windsurf](/docs/agents/windsurf) | Coding agent | A token | `AGENTS.md` | Not yet |
| [Gemini CLI](/docs/agents/gemini-cli) | Coding agent | Browser sign-in, or a token | `GEMINI.md` | Not yet |
| [GitHub Copilot](/docs/agents/github-copilot) | Coding agent | A token | `.github/copilot-instructions.md`, `AGENTS.md` | Not yet |
| [Lovable](/docs/agents/lovable) | App builder | Browser sign-in, or a token | `AGENTS.md` | Not yet |
| [Bolt](/docs/agents/bolt) | App builder | Browser sign-in, or a token | None | Not yet |
| [Replit](/docs/agents/replit) | App builder | Browser sign-in, or a token | `replit.md` | Not yet |
| [v0](/docs/agents/v0) | App builder | Browser sign-in, or a token | None | Not yet |

Using something else? Any client that speaks MCP over HTTP can connect: see [Other MCP clients](/docs/agents/other-mcp-clients).

## The server

```
https://api.stableship.ai/mcp
```

Most tools sign in through your browser: they open app.stableship.ai, you pick the workspace and approve, and there is no token to copy. Where a tool cannot, make a token in Stableship: **Settings**, then **Connections**, then **API tokens and connected apps**. You can connect your agent during setup, or later from **Settings**, then **Connections**.

## What to ask your agent

Once it is connected, plain requests work:

- "Fix the blocking Stableship findings, one at a time, and verify each fix."
- "What did Stableship find on the last release?"
- "Read the Stableship rules before you change the invoices table."
- "Run a Stableship check on this branch before I push."

No agent connected? Choose **Send to my agent** on a release: it gives you a private link to the fix plan that any agent or builder can read, for 7 days.