Guides

AI agents (MCP)

skiff mcp serves your projects over the Model Context Protocol, so an assistant in your editor or terminal can ship and debug them with you.

Connect an assistant

1. 1

   Create an API token

   In the dashboard, open Settings → API tokensand create one. It's shown once and acts for that team only.
2. 2

   Add Skiff as an MCP server

   Most clients take a block like this:

   mcp config

   {
     "mcpServers": {
       "skiff": {
         "command": "skiff",
         "args": ["mcp"],
         "env": {
           "SKIFF_URL": "https://dash.example.com",
           "SKIFF_TOKEN": "skiff_..."
         }
       }
     }
   }

   From a terminal client, the same thing is usually one command:

   shell

   claude mcp add skiff \
     --env SKIFF_URL=https://dash.example.com \
     --env SKIFF_TOKEN=skiff_... \
     -- skiff mcp

3. 3

   Ask

   “Why is checkout throwing 500s since this morning?” — “Deploy the API and tell me if it comes up.” — “Roll the site back to yesterday's build.”

What it can do

• list_projects, get_project — state, URL, replicas and the last deploy.
• deploy — build the latest commit and wait for it, returning the end of the build log.
• list_deploys, deploy_log, rollback — history, build output, and putting an earlier build back without rebuilding.
• app_logs— the running app's recent output.
• app_traffic — requests, error rate, latency, top paths, error paths, the deploys in that window, and uptime incidents.
• get_env, set_env— read and change environment variables. Changes merge with what's there and apply on the next deploy.