Codex

OpenAI's Codex (the GPT-5-Codex–powered terminal and IDE assistant) supports MCP servers in its configuration. Connecting MindWiki gives Codex access to the same 20 tools available to other MCP clients — search, read, capture, write, propose, and more.

Codex is most often run as a local CLI / IDE extension where browser-based OAuth isn't always practical. The recommended path is a personal API key.

Setup with API key

  • Mint an API key at mindwiki.io/account/api-keys.
  • Choose scopes — read+write for full capability, read only for browse-only workflows.
  • Copy the key (it's shown exactly once).

Add MindWiki to your Codex MCP configuration. The exact location depends on your Codex version — typically a config file referenced from your home directory:

json
{
  "mcpServers": {
    "mindwiki": {
      "url": "https://api.mindwiki.io/mcp?token=YOUR_API_KEY"
    }
  }
}

Restart Codex to pick up the configuration.

Storing the key safely

  • Never commit the key to a repo.
  • Reference it from an environment variable like MINDWIKI_API_KEY.
  • Revoke and re-mint if a key ever leaks — see API Access.

Setup with OAuth

If your Codex version supports remote MCP / OAuth, the URL alone is enough:

json
{
  "mcpServers": {
    "mindwiki": {
      "url": "https://api.mindwiki.io/mcp"
    }
  }
}

The first invocation triggers a browser approval flow. After approval the connection is reused.

Verify it works

Inside a Codex session:

> Search my MindWiki for anything about CI failures and summarize

Codex calls mindwiki_search, returns the matches, and writes the summary inline.

> Capture this terminal output insight to my MindWiki: "CI flake 
  was a missing env var, not a real failure."

Codex calls mindwiki_capture. The capture lands in capture/.

What Codex is good at

Codex is built around code-aware tasks, so the highest-leverage MindWiki workflows tend to be:

  • Capture from CI / build context. Save useful diagnostic insights without leaving the terminal.
  • Read project notes during code work. "Read my MindWiki page about the auth refactor before suggesting an API change."
  • Decision records. Have Codex draft a decision page based on the trade-off you just hashed out.
  • Search for prior art. "Have I written about this performance pattern in my vault before?"

Permissions

The scope model is the same as everywhere else:

  • API key — scopes you selected at creation.
  • OAuth — full read+write.

For Codex running in CI or other automated contexts, mint a dedicated key with the narrowest scopes that work, name it after the workflow, and revoke when the job is done.

Troubleshooting

  • `TOKEN_INVALID` — the key is wrong or revoked. Re-check or re-mint.
  • `INSUFFICIENT_SCOPE` — the key doesn't have the required scope for the tool Codex tried. Mint a key with the missing scope.
  • Tool not found — confirm Codex is reading the correct config file and the url is exact.

Where to go next