Because it fronts Fastly's whole API, the same connection that reads your cache hit ratio can also purge cache and change service configuration. That is the point, but treat the key you hand it as production credentials — the README does not publish a per-tool list, so the safe read is that whatever your key can do, the assistant can ask for.
An MCP server that puts the Fastly API behind an AI assistant. Fastly's API covers everything the web UI does, so the reachable surface is broad: CDN service configuration, cache purges, WAF and TLS settings, real-time and historical stats, and VCL or Compute@Edge deployment. The assistant never sees your key — it talks to this local helper, which holds it.
- Your services listed, and the domains configured on any one of them
- Cache purged for a named service
- Traffic patterns and cache hit ratio read back over a time window
- Backend servers and other configuration read out of a service
- The wider Fastly API surface — WAF rules, TLS, VCL and Compute@Edge — reachable through the same connection
A Fastly account and an API key, passed as `FASTLY_API_KEY` in the client's env block. The Fastly CLI must be installed. The README's config runs it as `node path/to/fastly-mcp.mjs`, so you install from the repository rather than a published binary; the package is named `fastly-mcp-server` at 0.1.0 and the badges mark the project beta and Node.js 16+.
Build from source — clone the repository and build it, then point your client at the binary