@fyio/buildfyio-mcp

The official Model Context Protocol server for Buildfyio. Lets AI agents deploy, roll back, tail logs, diagnose build failures, and manage env vars on Buildfyio — with scoped tokens and human-in-the-loop safety for destructive ops.

Quick install

npm i -g @fyio/buildfyio-cli
buildfy login
buildfy mcp install claude-code    # or: claude-desktop | cursor | windsurf

Restart your agent host. Then try:

"List every project in my Buildfyio workspace."

"Deploy the main branch of agent-demo to production and tail the build logs until it's done."

"The last production deploy of agent-demo failed. Diagnose it and either fix the env var or retry."

Manual install (any MCP client)

Add to your client's mcp_servers.json (or equivalent):

{
  "mcpServers": {
    "buildfyio": {
      "command": "npx",
      "args": ["-y", "@fyio/buildfyio-mcp"],
      "env": {
        "BUILDFYIO_TOKEN": "bfy_...",
        "BUILDFYIO_ORG": "org_..."
      }
    }
  }
}

Mint a token at Settings → API Tokens or via buildfy mcp install. Tokens carry one or more scopes: read, deploy, write.

Tools

| Tool | Scope | What it does | | ----------------------- | ------- | ---------------------------------------------------------------------------- | | list_projects | read | List projects in an org | | create_project | write | Create a project from a git repo URL | | deploy | deploy | Trigger a deployment | | get_deployment_status | read | Status, URL, failure reason | | get_logs | read | build / deploy / runtime logs | | get_build_failure | read | Structured JSON diagnosis of a failed build (agent-native) | | set_env_var | write | Create or update an env var | | rollback | write | Roll production back. Destructive — requires confirm: true |

Safety model

• Scopes. Every tool declares a required scope. If the token doesn't have it, the server refuses with TOKEN_MISSING_SCOPE — an actionable error the agent can explain to the user.
• Confirmation. rollback requires confirm: true in its arguments. Agents that call it without confirming get a validation error.
• Audit. Every call is logged with actor_type = ai_agent so the dashboard clearly distinguishes agent activity from humans.
• Rate limits. 120 req/min per token by default. Returns HTTP 429 with Retry-After.

Remote endpoint

https://mcp.buildfyio.com/v1/mcp — OAuth 2.1 + PKCE.

Use this from Claude.ai web, custom in-house agents, or any client that supports Streamable-HTTP transport.

License

MIT © Buildfyio