Connect your AI

Authentication

There are two ways in — both end up as a revocable key on your account:

• Browser login (recommended). Clients that speak the MCP auth spec — Claude.ai and Claude Desktop connectors, Claude Code, MCP Inspector — just take the server URL. On first use they open a browser window where you sign in to Ontonym and approve access. No key to copy. The token it mints shows up under Settings → MCP keys and can be revoked there any time.
• MCP key. For clients that want a static header, mint a key in Settings → MCP keys and send it as Authorization: Bearer aio_…. Treat it like a password — it is shown once.

Claude (web & desktop)

1. Open Settings → Connectors in Claude (claude.ai or the desktop app) and choose Add custom connector.
2. Name it Ontonym and paste the server URL: https://mcp.ontonym.com/mcp
3. Click Connect — a browser window opens; sign in to Ontonym and approve. Done, no key needed.

On team/enterprise plans an admin may need to enable custom connectors first.

ChatGPT

ChatGPT doesn't take a static key — the browser login is the only path, and it works out of the box. Custom MCP servers live in the unified Plugins directory (paid plans; on Team/Enterprise an admin may need to allow it):

1. Enable developer mode: Settings → Security and login → Developer mode (the “elevated risk” toggle — it allows unverified servers).
2. Open Settings → Plugins and hit the + button (New Plugin). Name it Ontonym, set Connection: Server URL to https://mcp.ontonym.com/mcp, pick OAuth as authentication and confirm the risk checkbox.
3. ChatGPT opens the Ontonym sign-in — log in and approve. The token it mints appears under Settings → MCP keys and is revocable there.
4. In a chat, click + in the composer, pick Ontonym, and approve the connection on first use.

Claude Code

Add the server, then authenticate from inside the session:

claude mcp add --transport http ontonym https://mcp.ontonym.com/mcp

In a Claude Code session, run /mcp and pick Authenticate next to ontonym — it opens the browser login. Prefer a static key instead?

claude mcp add --transport http ontonym https://mcp.ontonym.com/mcp \
  --header "Authorization: Bearer aio_your_key"

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json in a project, with an MCP key:

{
  "mcpServers": {
    "ontonym": {
      "url": "https://mcp.ontonym.com/mcp",
      "headers": { "Authorization": "Bearer aio_your_key" }
    }
  }
}

Recent Cursor versions can also connect without the headers block and offer a browser login instead.

VS Code

For Copilot agent mode, add .vscode/mcp.json to your workspace — VS Code prompts for the browser login on first use:

{
  "servers": {
    "ontonym": {
      "type": "http",
      "url": "https://mcp.ontonym.com/mcp"
    }
  }
}

Other clients

Any client that supports remote Streamable-HTTP MCP servers works with the URL plus the key header:

{
  "mcpServers": {
    "ontonym": {
      "url": "https://mcp.ontonym.com/mcp",
      "headers": { "Authorization": "Bearer aio_your_key" }
    }
  }
}

Stdio-only clients can bridge with mcp-remote:

{
  "mcpServers": {
    "ontonym": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.ontonym.com/mcp",
               "--header", "Authorization: Bearer aio_your_key"]
    }
  }
}

Using it

You don't call tools yourself — just talk to your AI. Some prompts to start with:

• “List my Ontonym memories.”
• “Save this conversation into my startups memory.”
• “Search the startupsmemory: what do we know about vector databases?”
• “Show the latest activity in apache-airflow.”

Under the hood the client reads the memory's extraction prompt, does the extraction on your model, and submits the structured result — your raw text never runs through our LLM. Reading works on any memory you belong to (plus public ones); writing requires membership.

No account yet? Register free — personal use is free without limits; a paid plan is only needed to share a memory with someone else.