How do I connect local Codex to the right Montage workspace?
Codex can connect to Montage as a remote MCP server over Streamable HTTP. Its CLI and IDE share MCP configuration, so choose the account and configuration scope deliberately before connecting another coding surface. This guide concerns local Codex tooling, not ChatGPT web app packaging. It describes documented setup rather than a live-certified partnership, and its first verification requires no collection job or model-driven research run.
[mcp_servers.montage]
url = 'https://app.montages.co/mcp'codex mcp add montage --url https://app.montages.co/mcp
codex mcp login montageCall Montage signals_status only. Report the returned workspace and source coverage; if authorization fails, stop and report the error.Locate the shared configuration
Inspect your existing ~/.codex/config.toml before adding a server. Preserve unrelated MCP entries and check whether a montage name already exists. Because the CLI and IDE share this configuration, changing its URL can affect both interfaces. Use the exact HTTPS endpoint rather than a ChatGPT connector manifest or a local executable path.
Register the Streamable HTTP endpoint
Use codex mcp add montage --url https://app.montages.co/mcp, or add the TOML table shown below through your normal configuration workflow. These are alternative registration methods, not two required installations. Registration records a remote server address; it does not itself grant workspace access or prove that tools are authorized.
Complete the documented OAuth flow
Run codex mcp login montage when you are ready to authorize the client. Codex normally opens a browser for OAuth. Review the Montage identity and requested access before approving, then return to the same Codex environment. Normal OAuth setup does not require a Montage API key, and no credential belongs in a prompt or committed TOML file.
Confirm the intended coding surface
Return to the CLI or IDE that will use the connection and inspect its available MCP tools. If the IDE was already open, follow its normal configuration refresh process rather than creating another server with a slightly different name. Keep initial tool permissions narrow; connecting a coding assistant does not authorize unrelated workspace changes.
Finish with authorized signals_status
Ask Codex to call Montage signals_status and inspect the actual response for the intended workspace and available coverage. Do not treat a saved TOML file or a tool listing as equivalent proof. If authorization fails, use Codex's login controls again with the correct account; if transport fails, check the HTTPS URL and installed version. Stop without collection or writes until this read-only check succeeds.
Keep the boundaries clear.
- Documented local Codex setup, not a ChatGPT web plugin or live interoperability certification.
- CLI and IDE share MCP configuration; avoid unintentionally replacing an existing server.
- OAuth access and workspace limits remain governed by the signed-in Montage account.
Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.
Primary references
Documentation reviewed 2026-11-13. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.