Skip to content

Connect Codex to Sensefold

Part of Sensefold for agents — the overview, permission tiers, and tool table.

Codex supports remote MCP servers with OAuth. Use OAuth for normal interactive setup: there is no Agent key to create, copy, or store.

The Codex app, Codex CLI, and IDE extension share MCP configuration on the same machine, so one connection is available in all three.

  1. Open Settings → MCP servers → Add server.
  2. Enter Sensefold as the name, choose Streamable HTTP, and paste:
    https://api.sensefold.app/mcp
  3. Save the server, then restart Codex.
  4. Select Authenticate next to Sensefold. Sign in in the browser and choose a permission tier. Start with read-only unless Codex needs to save or update items.
  5. Type /mcp in the composer to confirm that Sensefold is connected.

Run:

Terminal window
codex mcp add sensefold --url https://api.sensefold.app/mcp
codex mcp login sensefold

Complete the browser authorization, then run codex mcp get sensefold — it prints the server with transport: streamable_http and the URL above — or use /mcp in the Codex terminal UI.

For a manual configuration, add this to ~/.codex/config.toml and then run codex mcp login sensefold:

[mcp_servers.sensefold]
url = "https://api.sensefold.app/mcp"

Codex’s own approval settings decide whether it asks before running a tool; Sensefold’s permission tier decides which tools exist at all.

Use a manual Agent key only for a headless environment that cannot complete a browser OAuth flow. Create it in Sensefold → You → For agents, keep the secret in an environment variable, and point Codex at that variable instead of writing the key into a file or chat message:

Terminal window
codex mcp add sensefold --url https://api.sensefold.app/mcp \
--bearer-token-env-var SENSEFOLD_AGENT_KEY
  • “Before implementing, search my Sensefold library for the API design notes.”
  • “Save this investigation summary as a note.”

An agent-readable version of this setup lives at /for-agents/skill.md — paste that URL to Codex and let it guide the setup. See agent permissions for tiers and undo, and troubleshooting if the login does not complete.

Tiers, revocation, and undo: Permissions. What the client receives: Privacy for connected AI. Problems: Troubleshooting.