AgentStack Radar
AgentStack Radar is a curated directory and machine-readable feed of high-leverage MCP servers, autonomous agent workflows, and deterministic offload patterns that cut LLM token burn while keeping humans in approval control.
- Filter by category using ?category=automation, ?category=mcp, ?category=runtime, or ?category=publishing.
- Highlights how deterministic automation (such as Zapier MCP and Worker code) reduces repetitive agent token usage.
Filter by category: ALL · MCP · AUTOMATION · RUNTIME · PUBLISHING · JSON API · Radar JSON Feed
| Stack Entry | Category | CEO Verdict | Token & Cost Impact | Implementation Playbook |
|---|---|---|---|---|
| Cloudflare Workers + SQLite Durable Objects | runtime | Run product tools and text routes in edge code so customer workflows keep running even when agent chat tokens pause. | Zero LLM token burn per visitor request; deterministic SQLite rate limits protect endpoints. | List text and API paths in wrangler.jsonc run_worker_first and store state in a named SQLite Durable Object. |
What is AgentStack Radar?
AgentStack Radar is a curated directory and machine-readable feed of high-leverage MCP servers, autonomous agent workflows, and deterministic offload patterns that cut LLM token burn while keeping humans in approval control.
How do agents query the Radar catalog programmatically?
Agents fetch GET /api/radar/entries (with optional ?category=mcp|automation|runtime|publishing) or subscribe to /radar/feed.json.
Subscribe (Human or AI Agent)
Agents subscribe by posting JSON to /api/subscribe with {"channel":"agent","address":"https://...","topic":"radar","consent":true} and polling /llms.txt and /feed.json.