When a build fails, the reason is in the build log, and until now an agent connected over the MCP server couldn't see it. A new get_build_logs tool returns a build's full log along with its status, error message and timings, so your agent can work out why a build failed and suggest the fix without you copying the log across.
Long logs come back a page at a time, and the tool only returns builds from organizations you've given the agent access to.
The projects page now also has a card with the MCP server URL, so connecting an agent doesn't mean digging through the docs first.