Stop runaway MCP agents. Team gateway between Cursor/Claude and remote MCP servers — hard token budgets, kill switch, auto-kill loops. One Docker command. Self-host, one-time license — no subscription.
We built this after watching agents burn through tokens. TokenControlPlane is a self-hosted gateway between your AI clients (Cursor, Claude Desktop, Windsurf) and your MCP servers / OpenAI-compatible model APIs.
What you get:
Hard monthly token budgets (per key, per server)
A kill switch — flip it and clients get a 402 on the next request
Auto-kill on runaway loops (>120 req/60s per key)
A dashboard that actually shows usage
One static Go binary, or one Docker command
docker run -p 8080:8080 -v mcp-data:/data ghcr.io/devthinker-ai/tokencontrolplane:latest
We built this after watching agents burn through tokens. TokenControlPlane is a self-hosted gateway between your AI clients (Cursor, Claude Desktop, Windsurf) and your MCP servers / OpenAI-compatible model APIs.
What you get:
Hard monthly token budgets (per key, per server)
A kill switch — flip it and clients get a 402 on the next request
Auto-kill on runaway loops (>120 req/60s per key)
A dashboard that actually shows usage
One static Go binary, or one Docker command
docker run -p 8080:8080 -v mcp-data:/data ghcr.io/devthinker-ai/tokencontrolplane:latest
Measured overhead: +0.04ms p50.