v3.9.0

coolify-mcp 3.9.0 is a release of fixes. The biggest one is that its calls to Coolify no longer follow redirects. Alongside it, a hung instance can't stall a fleet, log requests get a ceiling, and errors tell you how to diagnose them.

Here's the bug. With Cloudflare Access in front of Coolify, a call could be redirected to the Access login page. Fetch followed it, and the client handed back that HTML page as the endpoint's result. Now the call refuses the redirect, and the error names Cloudflare Access and the service token that fixes it.

Refusing redirects also protects a credential. On a redirect to another host, fetch strips only the Authorization header. Your custom headers, the Access service token among them, would go along to that host. Now they stay with Coolify, and HTTP mode's token check stops following redirects too.

An instance that accepts the connection and never answers used to hold list_instances and the resource listing for fetch's 300 second timeout. Both now give each instance 10 seconds, then abort. The slow instance is reported as no answer within 10 seconds, and the listing still returns the rest.

Log requests get a ceiling. On logs and application logs, lines is now a whole number from 1 to 10,000, and the schema shows the model that limit. It used to take any number, and Coolify reads minus one as every line, so one call could pull a whole container log into context.

If you run several instances, each COOLIFY_INSTANCES entry now gets the same startup checks as the single instance settings. Errors name the entry by index and name, all reported at once. Doctor shows them on a fleet config line of their own, apart from the default instance's config check.

When Coolify is unreachable, or the token is refused with a 401 or a 403, the error now ends with this sentence. It says this server's environment because in HTTP mode, the person reading the error didn't configure the server. Other errors are unchanged.

One change can break a working setup. If your http base URL redirects to https, that used to work. Now every call fails with this message, and any other redirect fails naming its host. The fix is one line: use the https address. Doctor now fails a redirect too, where it used to warn.

Before you upgrade, run the 3.9.0 doctor with your server's environment. If reachability fails on a redirect, fix the base URL first. Then install 3.9.0 from npm, or drag the new extension file into Claude Desktop. From then on, a Cloudflare Access login page comes back as an error you can act on.