StackTrackerdocs
AI layer (MCP)

MCP overview

What the StackTracker MCP server is and what agents can do with it.

StackTracker ships an MCP server so AI agents can read and change your workspace directly. It is a JSON-RPC endpoint at https://stacktracker.tools/api/mcp/mcp, built with the mcp-handler package (createMcpHandler), and it responds to GET, POST, and DELETE.

Agents call named "tools". Each tool maps to one action against your StackTracker workspace, which is backed by Supabase.

The server exposes 20 tools, split by what they do:

  • 9 read tools. These return data and change nothing. Safe to call.
  • 11 write tools. These mutate your workspace: projects, costs, subscriptions, time, revenue, and more.

Every write tool revalidates cached data right after it runs, so your dashboard reflects agent actions immediately.

The data an agent sees is the same workspace your dashboard shows. There is no separate copy. A subscription an agent adds shows up on your dashboard, and a project you create in the dashboard is visible to the agent.

Client setup

Point Claude Code, ChatGPT, Cursor, Hermes, or OpenClaw at the server.

Authentication

How the Bearer token works and how to keep it safe.

Tool reference

All 20 tools, their params, and worked JSON-RPC examples.

Concepts and glossary

The entities StackTracker tracks and the metrics it uses to score a project.

Client setup

Point your MCP client at the StackTracker server.