Liminality · Model Context Protocol

Solve anything,
over MCP.

Type a request. We run it and hand back the answer.

first 3 solves free · connect any MCP client · no install

The algorithm

We build context-dependent determinism for your AI.

Liminality turns a request into an algorithm. First it reduces the request to the sub-questions that actually decide the answer. Then it pins the variables that matter for your situation and matches each piece to a tool that can run it. Only then does it execute. Whatever route runs gets stored with the verdict it earned, so the same situation produces the same route next time.

In plain English Ask the same thing in the same situation, get the same answer, the same way. Not a new guess every time you call. Your context picks the steps, and the steps hold.

What the MCP part means

MCP, the Model Context Protocol, is a shared standard for letting an AI tool talk to data and actions that live somewhere else. Connect once and the assistant can read a file, call an API, or run a step, with no custom integration for every pair. USB did this for hardware. MCP does it for what an AI can touch.

Liminality is one of those connections. It is not a pile of tools. It is a single endpoint that takes a request and returns a solved route, so any MCP client you already use can hand it the hard part.

From email to your first solve in a few minutes.

01
Create your account

Just an email and a password. You get a key on the spot, no plan to pick and no card.
02
Connect the MCP

Paste one line into your client. We send you straight back to where you were working.
03
Solve

Hand Liminality a request. It reduces it, grounds it, runs it, and learns from it.
04
Want more solves? Buy a pack

The first 3 solves on your key are free. Pick a pack only when you need more. Checkout is Stripe.

Get your key See how connecting works →

Pricing

One solve is one request, start to finish. The back-and-forth inside a solve never bills.

Free to start

$0 / 3 solves on every key

3 solves on any new key
The full engine, no feature gates
No card needed

Start free

Starter

$5 / 20 solves

25 cents a solve, start to finish
Follow-ups inside a solve are free
Packs never expire

Get the starter pack

Connect Liminality to your AI client

It takes a couple of minutes. Get a key, point your client at the endpoint, start solving. The handshake and the tool list are free — only a solve counts against your pack.

Endpoint https://liminality.physea.ai/mcp Remote HTTP transport. Auth header: X-API-Key. Get a key →

Claude Code

Grab your key at signup (first 3 solves free).
Run this in your terminal:
claude mcp add liminality --transport http https://liminality.physea.ai/mcp --header "X-API-Key: YOUR_KEY"
Ask your agent to use Liminality. That's it.

Cursor

Click to install, then paste your key into the X-API-Key header:
Add to Cursor

Or add it by hand in ~/.cursor/mcp.json:

{
  "mcpServers": {
    "liminality": {
      "url": "https://liminality.physea.ai/mcp",
      "headers": {
        "X-API-Key": "YOUR_KEY"
      }
    }
  }
}

VS Code

One-click install, then paste your key when prompted:
Add to VS Code

Or add it by hand in .vscode/mcp.json:

{
  "servers": {
    "liminality": {
      "type": "http",
      "url": "https://liminality.physea.ai/mcp",
      "headers": {
        "X-API-Key": "YOUR_KEY"
      }
    }
  }
}

Any other MCP client

Add a remote / HTTP MCP server pointing at https://liminality.physea.ai/mcp.
Set the header X-API-Key to your key.

ChatGPT, Claude & Perplexity OAuth

Open your client's connector settings:
- Claude (Desktop or web): Settings → Connectors → Add custom connector.
- ChatGPT (Pro / Business / Enterprise): Settings → Connectors → Add (enable developer mode if prompted).
- Perplexity: Settings → Connectors → Add.
Paste the endpoint https://liminality.physea.ai/mcp and connect.
Approve the sign-in when the window opens — no key to paste. Your first 3 solves are free.

Get your key Ask us about an enterprise contract →

What a solve actually does

01
Hand it a messy request. It returns the sub-questions, the parts you actually have to answer, ranked.
02
It scores the real choices, then grounds the winner against endpoints that can run today.
03
It runs the route, reports whether it held, and stores it. The next person's solve beats yours.

Solve anything,over MCP.