Home/Docs/Get started

Connect an AI app

One connector for Claude, ChatGPT, Claude Code, Codex, Cursor, VS Code, Antigravity, Kiro, Windsurf, Gemini CLI and any other MCP client.

The connector address

https://api.wrenship.com/mcp

It's a remote MCP server (Streamable HTTP). There's no key to copy: when you add it, your AI app opens a Wrenship sign-in page, you pick the workspace it may use, and it receives its own access token. You'll find the same address and a copy button on the AI page of the dashboard.

Claude

Open Settings → Connectors → Add custom connector.
Paste the address and save.
Press Connect, sign in to Wrenship and choose a workspace.

ChatGPT

Open Settings → Apps & Connectors → Advanced and turn on developer mode, then Create.
Paste the address and save.
Sign in to Wrenship when ChatGPT asks.

Claude Code

claude mcp add --transport http wrenship https://api.wrenship.com/mcp

Then run /mcp in Claude Code and sign in.

Codex

codex mcp add wrenship --url https://api.wrenship.com/mcp
codex mcp login wrenship

Cursor

Add Wrenship to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "wrenship": { "url": "https://api.wrenship.com/mcp" }
  }
}

Cursor opens the sign-in page the first time it uses a Wrenship tool.

VS Code

With GitHub Copilot in agent mode, add .vscode/mcp.json (or use Command Palette → MCP: Add Server → HTTP), then press Start above the server:

{
  "servers": {
    "wrenship": { "type": "http", "url": "https://api.wrenship.com/mcp" }
  }
}

Antigravity

In the agent panel, open ⋯ → MCP Servers → Manage MCP Servers → View raw config and add:

{
  "mcpServers": {
    "wrenship": { "serverUrl": "https://api.wrenship.com/mcp" }
  }
}

Kiro

Open the Kiro panel, then MCP Servers → Open config, or edit ~/.kiro/settings/mcp.json:

{
  "mcpServers": {
    "wrenship": { "url": "https://api.wrenship.com/mcp" }
  }
}

Windsurf

In Cascade, open MCP servers → Manage → View raw config (~/.codeium/windsurf/mcp_config.json) and add:

{
  "mcpServers": {
    "wrenship": { "serverUrl": "https://api.wrenship.com/mcp" }
  }
}

Gemini CLI

Add to ~/.gemini/settings.json, then run /mcp auth wrenship:

{
  "mcpServers": {
    "wrenship": { "httpUrl": "https://api.wrenship.com/mcp" }
  }
}

The dashboard's AI page has each of these with the address filled in and a copy button.

Other MCP clients

Any client that supports remote MCP servers with OAuth works. Wrenship publishes standard discovery documents, so the client finds the sign-in and token endpoints by itself, and registers itself without a client ID.

An agent that can run terminal commands can also deploy without MCP, with the CLI: npx wrenship deploy --json.

What your AI can do

Tool	What it does
`list_projects`	Lists the workspace's projects with their live URL and latest build
`get_project`	Status of one project: live URL, production build, latest build, domains
`deploy`	Builds and publishes files, or redeploys a Git project. `environment: "preview"` gives a temporary URL kept for 7 days. For a new project, `site_domain: "wrenship.space"` picks that free address instead of `.wrenship.site`
`get_deployment`	Follows a build until it's live or failed
`get_logs`	Build log lines, optionally filtered
`rollback`	Puts an earlier successful build live again
`list_env`	Names and scopes of environment variables. Secret values are never returned
`set_env`	Creates or updates one variable, used from the next deploy
`add_domain`	Connects a domain and returns the DNS record to add
`check_domain`	Re-checks DNS and SSL for a domain
`get_analytics`	Requests, page views, p95 response time, errors and top pages

Staying in control

Rules per workspace. On the AI page, owners choose whether AI apps may deploy to production (never, with approval, or freely), deploy previews, roll back, and change environment variables.
Approvals. With approval on, a production deploy from an AI app waits for a person. The AI is told it's waiting and gets a link; you approve or decline in the dashboard. A request expires after an hour.
Activity. Everything an AI app does is in the activity log, marked with the app's name.
Disconnect. Remove an app on the AI page. It loses access at once and its pending requests are cancelled.
Roles. An AI app can never do more than the person who connected it.