Hey everyone,
We just open-sourced the Fastly Agent Toolkit. It is a collection of skills that teach AI coding agents how to work with Fastly.
If you’ve used the Fastly MCP Server, you already gave your agent the ability to act on Fastly. Skills are the other half: they give your agent the knowledge to get things right. Correct VCL patterns, the right CLI flags, proper deploy sequences - the kind of Fastly-specific expertise that models don’t pick up from their training data.
What’s in the toolkit:
- 6 skills covering the Fastly platform, CLI, VCL (Falco + XVCL), and local Compute testing (Viceroy + Fastlike)
- Works with Claude Code, Gemini CLI, Codex, Cursor, Amp, Cline, and any agent that reads .agents/skills/
- Plain markdown files, zero lock-in, easy to customize
Does it actually work?
We ran A/B evaluations across 7 models and 84 test runs. Skills delivered a 1.6x wall-clock speedup, 30-70% fewer tool calls, and won 7 out of 8 tasks against baseline. The biggest gains show up in CDN configuration, VCL development, and Compute deploys - exactly where Fastly-specific knowledge matters most.
Get started:
npx skills add github:fastly/fastly-agent-toolkit --skill fastly --skill viceroy
Full install instructions for every supported agent are in the README.
We want to hear from you. If you’re using AI agents with Fastly today, tell us how it’s going and what tasks you wish your agent handled better. Try the skills and let us know what changes. Better results, fewer mistakes, faster workflows? If you’ve found a pattern that should be in the toolkit, PRs are welcome.
This is an open project and we want to build it with you. Drop a comment and tell us how you’re using agents with Fastly.