Agents can do real work on an Astro site. How well that goes depends on the context you give them and the checks around them. Most of this is useful for human developers too.
Tell agents how the repo works
Add an AGENTS.md to the root of the repo. Keep it short and specific:
- How to install, run, test and build.
- Where things live, and what not to touch.
- The rules a new team member would need, such as copy style or design tokens.
Tools that read CLAUDE.md can use the same content.
Give them a way to check their work
An agent that can’t verify a change is guessing. The minimum is a build that fails loudly:
astro check && astro buildAdd tests for anything that matters, and run them in CI so that every pull request is checked the same way.
Connect the tools they need
The Model Context Protocol lets an agent reach outside the repo. A .mcp.json file can declare servers for the project. EmDash goes further: every EmDash site has an MCP server, so agents can work with content under the same permissions as a person.
Keep people in charge
Have agents work on branches and open pull requests. Keep deploy keys out of reach. With EmDash, drafts can stay drafts until someone approves them.
Publish something agents can read
llms.txt is a proposed standard for a short, plain-text summary of a site, written for AI tools. It’s a small file, and it’s easy to generate from the data your Astro site already has.
Let customers leave notes
Page feedback closes the loop. A note pinned to the exact spot on a page gives a person or an agent a far clearer brief than a ticket.
See where you stand
Em Region’s Agent Readiness check looks for these files on every repo in your fleet.