Runnel
How it worksAIPricing

Runnel for AI assistants

Your assistant can mint private links, list them, revoke them, and check your quota. It works over MCP, the open protocol AI apps use to call tools. Four tools: mint_link, list_links, revoke_link, get_usage. Everything it mints counts against your plan's normal limits.

Remote: claude.ai, ChatGPT, and other MCP apps

Connect once from the app's settings. No install and no API key. Runnel asks you to approve access in your browser, signed in as you.

  1. In claude.ai: Settings → Connectors → Add custom connector. In ChatGPT: enable Developer mode in settings, then add an MCP connector.
  2. Paste https://app.runnel.link/mcp and confirm.
  3. The app sends you to Runnel to approve access. Approving grants minting, listing, revoking, and usage (metadata only).
  4. To let the assistant also read where your links point, tick "Allow reading destinations" on that page. It asks for your unlock key or passkey: the reading key exists only after you unlock.

If you tick the box, your reading key is sealed into that grant, encrypted to that grant's own token. Only requests that carry the token can use it. We store token hashes, not tokens, so the stored key is unreadable at rest, even by us. Destinations are decrypted per request, in memory, and never stored. Skip the box and the assistant sees metadata only: ids, labels, status, use counts.

Every connected app is listed under Connected apps on your API keys page. Revoke one and its access, including destination reading, ends at its next request.

Local: Claude Code, Cursor, VS Code, Claude Desktop

The local server runs on your machine and talks to Runnel with an API key. Create one on the API keys page, then add:

{ "command": "npx", "args": ["-y", "@runnel-link/cli", "mcp"], "env": { "RUNNEL_API_KEY": "rnl_..." } }

In Claude Code it is one line:

claude mcp add runnel -e RUNNEL_API_KEY=rnl_... -- npx -y @runnel-link/cli mcp

To let it read destinations, pair the machine once. Click Show pairing key on the API keys page, run runnel auth pair, and give the key the reveal permission. Decryption then happens on your machine. The server still stores only ciphertext.

CLI and Claude Code skill

The same package is a CLI: runnel mint, runnel list, runnel revoke, runnel usage, with --json for scripts. runnel skill install adds a Claude Code skill that teaches the agent when and how to use it.

What your AI can and cannot see

Always

Link ids, labels, status, use counts, expiry, your plan usage.

Only with your say-so

Destinations. Remote apps need the consent checkbox, and local machines need pairing. Both are revocable, and the server itself keeps only ciphertext at rest.

MCP is included on every plan, Free included. Calls draw on the same quotas as the dashboard: links per month, active links, proxied bandwidth. If automated abuse ever forces it, we may add per-plan request ceilings. We will say so before we do. See plan limits.