Redirect_uri mismatch for remote server with OAuth authentification on 2nd start when default port is taken

[fredericbarthelet]

Issue

Users see redirect_uri mismatch OAuth error when restarting their MCP client.

What's happening

From the information I managed to gather, here is what I think is happening :

• port 3334 is used by another process on the machine, a random port is chosen to start the callback server
• the only location the port is written to is the lockfile (together with pid)
• a new OAuth client is registered with a callback uri including the random port
• first auth is happening without any issue
• at first MCP client shutdown, lockfile is cleaned by cleanupHandler process
• when restarted, the MCP client uses port 3334 if free, or chose another random port
• there is a redirect_uri mismatch error because the port in the redirect URL used during client creation differs from the one used at the moment

Temporary solution

In the meantime, I'll add a less frequently used port in the config to increase chances request port is actually used all the time by mcp-remote.

Long term solution

Here is my proposal, happy to submit a PR if you're fine with it :) :

• add port to client_info.json (or keep it as it is, and parse the first URL in redirect_uris to extract port information)
• use this value as port target value for callback server starting
• on server started, check that this port is actually the one being used by the server (it did not fallback to another port) - exit early if it does, not a chance the process will succeed

[geelen]

OH MY GOD that makes so much sense. I have been so confused by this error, THANK YOU for figuring it out.

Please send a PR if you're able, otherwise I'll jump in and do the quick fix tomorrow.

[dcramer]

this is an everyday bug fwiw - hits non-stop for folks. Was super confused til I found the "find unused port" code" :)

[geelen]

Fixed in v0.1.5. I think this should fix tons of the flakiness, but please let me know if not!