Install GNO for the OpenAI Codex CLI as an agent skill, an MCP server, or both. The skill path skips the MCP handshake. The agent calls the CLI it already knows, plus bundled recipes for lookup, capture with provenance, meeting ingestion, and citation discipline.

Connects via: Agent skill + MCP server
Requires: Bun 1.4.1+ and one indexed folder
Data boundary: Corpus and index stay on your machine

Install

Connect Codex in one command.

gno skill install --target codex

gno agents install --target codex

gno mcp install --target codex

gno skill show --file recipes/brain-first-lookup.md

What it unlocks

Retrieval without the MCP protocol layer
Bundled second-brain recipes the agent can follow
Capture with structured provenance straight from the agent
The same index every other GNO surface reads
Optional manual import of past Codex sessions into a separate, redacted archive index; nothing imports automatically unless you opt in to a daemon schedule

Prove it actually works

A written config file is not a working integration. GNO distinguishes the two, and so should you.

Run gno skill paths to see exactly where the skill was written, and gno agents verify to confirm the instruction block is current.
Ask the agent to search your notes for a phrase only your corpus contains.
GNO cannot prove a client loaded an installed skill, so the retrieval answer is the proof. The install step reports what it wrote, not what the client did with it.

Keep reading