Loop with AI MCP (loopwithai-mcp) server

Gives any MCP client (Claude Code, Claude Desktop, Antigravity, Cursor, VS Code…) native tools to work with a SyncEfforts project — read its rules/context/changes and write issues/tasks/docs/decisions. Stdio transport; authed by an org-scoped API key. No build step.

Setup

1. cd mcp && npm install
2. Generate an API key in the app: Org → API → Generate key. Optionally grab a project id (from list_projects or the project URL) to use as the default.
3. Add it to your client's MCP config:

{
  "mcpServers": {
    "syncefforts": {
      "command": "node",
      "args": ["/path/to/loopwithai/mcp/server.mjs"],
      "env": {
        "SYNCEFFORTS_API_KEY": "se_live_...",
        "SYNCEFFORTS_PROJECT_ID": "<optional default project id>"
      }
    }
  }
}

The config lives in your client's MCP settings, not in this file — this is just the snippet to paste. The block above (key mcpServers) is what Claude Code / Claude Desktop / Cursor / Antigravity use:

• Claude Code — .mcp.json at your repo root (block above), or: claude mcp add loopwithai --env SYNCEFFORTS_API_KEY=se_live_... --env SYNCEFFORTS_PROJECT_ID=<id> -- node /path/to/loopwithai/mcp/server.mjs

• Claude Desktop — claude_desktop_config.json → mcpServers.

• Cursor — .cursor/mcp.json → mcpServers.

• Antigravity — its MCP settings → mcpServers.

• VS Code — uses a different shape: a .vscode/mcp.json in your workspace with the key servers and an explicit "type": "stdio" (not mcpServers):

  {
    "servers": {
      "syncefforts": {
        "type": "stdio",
        "command": "node",
        "args": ["/path/to/loopwithai/mcp/server.mjs"],
        "env": { "SYNCEFFORTS_API_KEY": "se_live_...", "SYNCEFFORTS_PROJECT_ID": "<optional>" }
      }
    }
  }

  (Add via the Command Palette → "MCP: Add Server" → it writes this file.)

SYNCEFFORTS_API_URL overrides the API base (defaults to the deployed se-api).

What the client gets

Tools

• whoami · list_projects
• get_context — bootstrap: rules (follow them) + open tasks + recent decisions + docs + change cursor
• list_changes(since) — sync feed
• list_issues|tasks|rules|decisions|docs|tags
• create_issue|create_task|create_doc|create_decision

Resources (when SYNCEFFORTS_PROJECT_ID is set): syncefforts://rules, syncefforts://context — loadable context.

Prompt: adopt_project_rules — loads the project's rules + open tasks and tells the agent to treat the rules as binding.

Governance

By design there is no create_rule/update_rule tool — agents read rules and record work (issues/tasks); humans author rules. An agent should surface rule proposals as an issue. Every write via this server is logged in the app's Activity feed as an AI action, attributed to this key's title.