Docs › Connectors
Use AstraCode from Claude, ChatGPT and other apps
Start, follow and stop AstraCode agent runs on your GitHub repositories from Claude, ChatGPT or any app that connects to remote MCP servers.
AstraCode is also an MCP server, at https://astracode.io/mcp. Connect it to Claude, ChatGPT or another app that supports remote MCP servers with OAuth, and you can ask that app to start an AstraCode agent on one of your GitHub repositories, follow it, stop it, and check your usage. The agent runs in the repository's own GitHub Actions, as a remote agent does, and its work comes back as a pull request.
Before you connect
- An AstraCode account with a verified email address.
- GitHub connected to your organisation: Integrations in your dashboard, Connected accounts, Connect, then choose the repositories AstraCode may see. An organisation admin does this once. An agent can run only on those repositories.
- The agent workflow in each repository you want to use. If it is missing, ask the app to set the repository up: AstraCode opens a pull request that adds
.github/workflows/astracode-agent.yml, and nothing changes until someone merges it.
Connecting
Add a connector with the address https://astracode.io/mcp (in Claude, from the directory or as a custom connector; in ChatGPT, from its apps). The app sends you to astracode.io to sign in, and a page there shows which app is asking, which account it will act for, and what it will be able to do. Allow connects it; nothing is shared until you do. The app never sees your password.
What it can do
| Tool | What it does |
|---|---|
| list_repositories | Lists the repositories your organisation connected, the only ones an agent can start on. |
| start_agent_run | Starts the agent on a repository with your task, from its default branch or one you name. It works on a new branch, `astracode/<run id>`, never pushes to an existing one, and opens a pull request when it changes something. Stopped after 60 minutes unless you ask for another limit, up to 360. |
| get_agent_run | How a run is going: status, steps taken, the last tool it used, and when it ends, its report and pull request. |
| list_agent_runs | The runs you started from the app, newest first. |
| stop_agent_run | Cancels a running run you started from the app. Work it has not pushed is lost; a pull request it already opened stays open. |
| set_up_repository | Opens a pull request that adds or updates the agent workflow. |
| get_usage | How much of your plan's allowance is used this period, and when it resets. |
The agent sees only the task you give it, not the conversation it came from, so write the whole task: "In acme/api, fix the failing test in tests/test_dates.py and open a pull request." Model use comes from your plan, as it does in the editor (pricing); the Actions minutes are the repository's own.
Disconnecting
Disconnect in the app, or under Connected apps on your dashboard's Integrations page, which lists every app connected to your account, what it can do and when it was last used. Disconnecting stops the next request at once, not when the current token runs out. A connection left unused for 30 days stops working and has to be connected again.
What is kept
Which app you connected, what you allowed, and when it was connected and last used, until you disconnect it. For each run started from an app, the repository and when; the task you gave it is not stored. A run's progress (steps, tool names, the report and the pull request link) is kept for seven days. The privacy notice has the rest.
If something goes wrong
- The repository is not connected: add it to AstraCode's GitHub access on the Integrations page, then ask again.
- The repository is not set up: ask the app to set it up, and merge the pull request it opens.
- The app says the connection failed after you allowed it: start connecting again from the app and finish within a few minutes. A sign-in left open for a long time expires on the app's side.
- Anything else: support@astracode.io.
All documentation
Get started
Agent
- Agent overview
- Approvals and permissions
- Plan mode
- What the agent can do
- Running commands
- Browser tools
- Checkpoints and undo
- Infrastructure as code
- Running several agents at once
- Subagents and background tasks
- Remote agents in GitHub Actions