Where this stands

Surface	Status	Why
ChatGPT (connector)	Unproven	Not yet demonstrated. It needs OAuth discovery from the endpoint and a registered hosted callback, both in Fleet's readiness work.
Codex (CLI and IDE)	Unproven	The configuration below is documented in Fleet's own runbook, with the bearer supplied by a sign-in helper; an end-to-end run has not been recorded.
OpenAI directory domain verification	Planned	This site can serve `/.well-known/openai-apps-challenge` once a verification token is issued; it answers 404 until then. It counts for verification once the MCP host is `fleet.oracis.ai` or a subdomain of it.

Everything on this page describes how the connection is designed to work. The word Unproven means exactly that: nobody has completed it on the deployed service yet.

ChatGPT

Settings, then Connectors, then Create.
Name it Fleet and paste the endpoint below. Choose OAuth.
Connect.
ChatGPT reads the server's OAuth metadata and opens the Fleet consent page in a new window.
Sign in and read the consent.
Client, redirect host, your email, your organization, the scopes. Approve or deny.
Use it in a conversation.
Enable the connector for the chat; read tools run without confirmation, write tools ask first.

EndpointStreamable HTTP · OAuth 2.0 with PKCE

https://fleet-api-fleet-api.up.railway.app/mcp

Codex

Codex configures a streamable HTTP MCP server in ~/.codex/config.toml (or a project's .codex/config.toml) and sends a bearer token read from an environment variable. Fleet's sign-in helper runs the OAuth flow on your machine and places the token in that variable; the token never appears on your screen or in the file.

~/.codex/config.tomlfrom the Fleet runbook, section 10

[mcp_servers.fleet]
url = "https://fleet-api-fleet-api.up.railway.app/mcp"
bearer_token_env_var = "FLEET_ACCESS_TOKEN"
env_http_headers = { "x-fleet-organization" = "FLEET_ORGANIZATION_ID" }

The x-fleet-organization header is needed only when your token carries no organization claim. The sign-in helper and its options are part of the invitation you receive; they are not public downloads yet.

Use cases

Five prompts that exercise the read and write tools. Replace the angle-bracketed values with your own.

Check a validation before you merge

Promptvalidate_status · read

Check the status of Fleet validation <id> and tell me which lane failed and why.

Validate the exact commit you are about to ship

Promptvalidate_submit, then validate_status · write, then read

Submit commit <sha> of <owner>/<repo> to Fleet for validation, wait for the receipt, and summarise what it binds.

Ask what Fleet has learned about this kind of work

Promptlessons_for · read

Before I start, ask Fleet for lessons about "<task in one sentence>" in <owner>/<repo>.

Publish a receipt as a check on the pull request

Promptpublish_request · write

Request publication of Fleet receipt <id> so it appears as a check on the commit.

Cancel a job you no longer need

Promptjob_cancel, then job_status · write, then read

Cancel Fleet job <id> and confirm its final state.

What Fleet refuses, on purpose

A call with no token, an expired token, or a token issued to another client id: 401.
A job, validation or receipt in another organization: not found.
A budget outside 1 to 2,000 cents or beyond your organization's ceiling: refused, nothing admitted.
A commit that changes its own verification policy: policy_changed.

Troubleshooting

ChatGPT cannot complete the OAuth step: Expected today: see Where this stands.
Codex answers 401: FLEET_ACCESS_TOKEN is not set in the environment Codex runs in, or the 8-hour session expired. Run the sign-in helper again in that shell.
validator_unavailable (ref …): Nothing ran. Quote the ref to support.

The general list is on the documentation page.

Connect from ChatGPT and Codex