Setup guide

Connect AiAkiv to your AI

Your AI only needs to speak MCP. The easiest path is to let that AI do it. Give it the line below and it asks two things, then sets up for you. Prefer to do it by hand? Follow the per-app guide lower down.

1. Easiest way - let your AI do it

Give the AI you're using the sentence below. It asks how you'll use AiAkiv (personal / team) and which app you're on, then writes the config file for folder apps or walks you through the clicks for web / desktop apps.

read https://www.aiakiv.com/llms-setup.txt and set up AiAkiv for this app. Guide me kindly, step by step, as if I'm new to this.

2. Manual install

Endpoint

https://mcp.aiakiv.com/mcp

⚠ The URL must end with /mcp. Without it the connection may succeed but zero tools appear.

For the exact config with your real project name and key baked in, copy it from the console. Or replace <PROJECT-NAME> in the snippets below with your project name.

Per-folder clients pick a project via the X-K2G-Project header. App-wide connections carry no header and follow the current ‘Main’ project; you switch Main under ‘Manage projects’ in the console.

One app-wide

Apps that can't split projects by folder. Connect the single URL; to change project, hit ‘Switch Main’ under ‘Manage projects’ in the console.

Claude Web URL

Settings → Connectors → ‘+’ → Add custom connector → paste the URL above. (Pro/Max. On team/enterprise, owner only: Settings → Org settings → Connectors.)

Claude Desktop

App-wide = account (bottom-left) → Settings → Connectors → Add custom connector → paste the URL. Folder = create .mcp.json in the working folder and paste the JSON below.

File · .mcp.json

{
  "mcpServers": {
    "AiAkiv": {
      "type": "http",
      "url": "https://mcp.aiakiv.com/mcp",
      "headers": {
        "X-K2G-Project": "<PROJECT-NAME>"
      }
    }
  }
}

ChatGPT classic URL

① A connector cannot be created from the app. Open chatgpt.com in a browser and add it on the web page. ② Click your user (bottom-left) → Settings → Security and login → turn on ‘Developer mode’. This is only needed until AiAkiv is listed in OpenAI's plugin directory (as of 2026-08); if that is too much, use the ChatGPT card's folder+key method instead. ③ Settings → Apps (connectors) → create an app → paste the URL above. (Plus and up.) Once made, it applies account-wide, including the ChatGPT app. ④ Finally, ChatGPT enables a connector per conversation. In a new chat click + in the composer and pick AiAkiv (or use the “Try in chat” button on the plugin page, which opens a chat with it already on). Skip this and no tools appear in a new chat.

xAI / Grok URL

grok.com/connectors → New Connector → Custom → paste the URL above.

Antigravity API key

Agent panel ‘…’ → MCP Servers → Manage MCP Servers → View raw config → paste the JSON below into mcp_config.json. It reloads on save. (File: ~/.gemini/antigravity/mcp_config.json; on some versions ~/.gemini/config/mcp_config.json)

The remote key MUST be serverUrl. url / httpUrl / type are not recognized (strict schema). The config is app-wide (no per-folder split), so the KEY picks the project: issue a project-scoped API key in the console and put it in the Authorization header (the key pins the project, so Main does not apply).

File · mcp_config.json

{
  "mcpServers": {
    "AiAkiv": {
      "serverUrl": "https://mcp.aiakiv.com/mcp",
      "headers": {
        "Authorization": "Bearer <API-KEY>"
      }
    }
  }
}

Cursor

App-wide = Settings (Ctrl/Cmd+,) → Tools & MCP → + Add New MCP Server → set Transport to streamable-http → paste the URL. Folder = create a .cursor folder and paste the JSON into mcp.json.

File · .cursor/mcp.json

{
  "mcpServers": {
    "AiAkiv": {
      "type": "http",
      "url": "https://mcp.aiakiv.com/mcp",
      "headers": {
        "X-K2G-Project": "<PROJECT-NAME>"
      }
    }
  }
}

Per folder

A config file in the working folder ties that folder to a specific project.

Claude Code

Create a .mcp.json file in the working folder and paste the JSON below.

File · .mcp.json

{
  "mcpServers": {
    "AiAkiv": {
      "type": "http",
      "url": "https://mcp.aiakiv.com/mcp",
      "headers": {
        "X-K2G-Project": "<PROJECT-NAME>"
      }
    }
  }
}

ChatGPT

The ChatGPT app itself cannot use OAuth. A folder project is set up with a static API key instead. Issue a project-scoped API key in the console, copy the whole config, and paste it into config.toml. The project stays pinned no matter what your Main project is. After editing .codex/config.toml, save it and restart the app. You can use it alongside the ChatGPT classic connector and other ChatGPT static keys. Just keep the names from colliding.

OAuth does not attach on this app. Neither a custom connector nor folder binding (the X-K2G-Project header) takes hold. That is why it pins with the key instead of a header. You can keep the ChatGPT classic account connector alongside it, but give the two different server names so you can say which one to save through.

File · .codex/config.toml

[mcp_servers.AiAkiv-proj]
url = "https://mcp.aiakiv.com/mcp"
http_headers = { "Authorization" = "Bearer <API-KEY>" }

Gemini CLI API key

Create a .gemini folder and paste the API-key config into settings.json. Issue the key per project in the console (shown once).

Gemini CLI's login session does not persist. You must use the API-key method.

File · .gemini/settings.json

{
  "mcpServers": {
    "AiAkiv": {
      "httpUrl": "https://mcp.aiakiv.com/mcp",
      "headers": {
        "Authorization": "Bearer <API-KEY>"
      }
    }
  }
}

Before you connect

The URL must end with /mcp (or zero tools appear).
OAuth clients like Claude Code and Cursor require re-authenticating when you change the project (header). Keep a separate config per folder.
OpenAI ships two apps, ChatGPT and ChatGPT classic, and they attach differently. Create the classic connector on the web page (chatgpt.com), not in an app. The ChatGPT app cannot do OAuth, so it attaches with a project-pinned API key in the folder (see the ChatGPT card). You may keep both, but give them different server names.
Gemini CLI must use the API-key method.
If a project has a persona, web apps like ChatGPT / Claude Web won't reliably follow it on their own. Add one line to the app's Custom Instructions: “At the start, call AiAkiv's get_save_target and adopt the persona.”

Verify

Once connected, the tool list shows names like save_memory and search_memory. Call get_save_target to confirm which team, project and domain you're saving to. No tools showing? The URL is almost certainly missing the /mcp suffix.

Machine-readable version for AIs: /llms-setup.txt