Connect LiquidLM in Codex

Remote MCP OAuth for Codex CLI with version-specific compatibility notes.

Prerequisites

  • Codex CLI installed locally.
  • A LiquidLM account with indexed content to search.
  • Prefer OAuth for interactive use; reserve bearer tokens for headless CI.

Check your Codex version

LiquidLM OAuth passed on @openai/codex@0.142.5 (2026-07-26). Codex 0.145.0 reached the authorization callback but rejected it with Authorization server response missing required issuer — tracked upstream at github.com/openai/codex/issues/31573, not a LiquidLM misconfiguration.

  • If you use the pinned build, prefix every command in the next steps with npx -y @openai/codex@0.142.5.
  • Retest the current stable Codex release before claiming latest-version support.

Print the installed version

codex --version

On a newer Codex, run the pinned build instead of changing your install

npx -y @openai/codex@0.142.5 mcp list

Add the remote server

codex mcp add liquidlm --url https://mcp.liquidlm.com/v1/mcp

Sign in with OAuth

Complete the browser sign-in that this command opens:

codex mcp login liquidlm --scopes mcp,profile,email

Confirm the server is registered

  • Expect liquidlm enabled with Auth = OAuth.
  • Inside an interactive session, /mcp shows the same connection.
codex mcp list

Verify a search

Ask Codex for something you know is in your vault:

  • Inspect the tool invocation to confirm the call reached LiquidLM.
  • Codex stores the server definition and OAuth credentials locally — do not replace OAuth with a literal token unless the environment is headless.

Example prompt

Search LiquidLM for our onboarding checklist and quote the steps.

Headless alternative

For CI, create a custom PAT with the smallest mcp.* grants needed, scoped to the required vaults, and pass it through the environment:

  • Send the token as Authorization: Bearer on direct HTTP — never commit it.
  • Revoke the PAT when the job completes.
export LIQUIDLM_API_TOKEN="lqlm_..."

Assistants hub

CLI setup