SELECT A PLAN

You're In. Let's Wire Up Your Engine.

Connect Claude, ChatGPT, Gemini, Cursor, VS Code, and others by signing in with Haakon, or use a platform key. Either takes under a minute.

Sign In with Haakon

Add the Haakon server URL to your tool, then sign in with this same account and approve the connection. No key to copy or paste, and you can disconnect any time from the tool's connector settings.

Every connection uses the same URL, which has no key in it:

https://api.haakon.ai/mcp/haakon

Requires an active Haakon plan. If you see "An active Haakon plan is required", choose a plan on this page and you'll be sent back to finish connecting.

claude mcp add haakon --transport http https://api.haakon.ai/mcp/haakon

codex mcp add haakon --url https://api.haakon.ai/mcp/haakon

{
  "mcpServers": {
    "haakon": {
      "url": "https://api.haakon.ai/mcp/haakon"
    }
  }
}

{
  "servers": {
    "haakon": {
      "type": "http",
      "url": "https://api.haakon.ai/mcp/haakon"
    }
  }
}

Gemini and any tool without a sign-in prompt: use Option 2 below.

Connect with a Platform Key

For tools without a sign-in prompt (such as Gemini), or if you prefer a key. Step 1: generate a key. Step 2: paste it into the connection URL below. Treat the key like a password, since anyone with it can use your plan.

Add the Key URL to Your Tool

Replace YOUR_HAAKON_KEY with the key you generated above. It goes directly in the connection URL below, with no separate provider key required.

https://api.haakon.ai/mcp/haakon?key=YOUR_HAAKON_KEY

https://api.haakon.ai/mcp/haakon?key=YOUR_HAAKON_KEY

https://api.haakon.ai/mcp/haakon?key=YOUR_HAAKON_KEY

{
  "mcpServers": {
    "haakon": {
      "url": "https://api.haakon.ai/mcp/haakon?key=YOUR_HAAKON_KEY"
    }
  }
}

{
  "servers": {
    "haakon": {
      "type": "http",
      "url": "https://api.haakon.ai/mcp/haakon?key=YOUR_HAAKON_KEY"
    }
  }
}

Setup

Haakon doesn't choose or change which model your AI tool uses -- that's set wherever you're calling from (Claude, ChatGPT, Gemini, Cursor, VS Code, and others). Haakon's cognitive lenses extend whatever model is already selected there; they don't substitute for one.

• For low-cost, fast performance: keep a flash or fast model selected in your calling tool. A lens applied to a fast model is what gets you closer to frontier-level output at flash-model cost.
• For maximum quality regardless of cost: select a premium or frontier model in your calling tool. Haakon's lenses extend that model's reasoning further still.

Pre-Flight Feasibility Gate

Before any lens runs, Haakon scores whether your request can be completed reliably as written (0–100). If it can't, Haakon stops your AI from guessing: no invented prices, no made-up data, no code written against files it hasn't seen. You see exactly what is missing instead. The audit also estimates the size of the task in rough token bands: Small (under ~5k), Medium (~5k–30k), Large (~30k–150k) or Extra Large (150k+). These are rough estimates of the whole task; actual usage depends on your AI model, its tools, retries and how much context it already holds. The score is a readiness rubric across four pillars (ground-truth data, logic clarity, constraint validity, single-pass viability), 25 points each. It is a heuristic, not a calibrated probability.

What a halted request looks like

Ask for a 35-delta options trade on a stock without supplying the live option chain and the audit stops it, listing what to provide:

Completion Confidence Score: 20/100 (NEEDS USER DATA TO AVOID FICTION)
1. Goal Check & Verdict   2. What It Will Take to Research
3. What It Will Take to Execute   4. Missing Inputs: the live option chain, current price, expiry

Option A: provide the missing inputs.
Option B: re-run with "feasibility_off" to proceed on stated assumptions.

How to use it

• Do nothing: the gate runs on every haakon call. Well-formed requests pass in a single turn.
• See the audit on demand: ["feasibility"].
• Proceed past a halt on assumptions: add "feasibility_off", for example ["feasibility_off", "auto_ultimate"].
• Coding requests (code_ultimate or any code_ lens) are advisory: the score and possible gaps are shown, but the request is never halted, so coding tools don't need the override.
• Pass the real task in your query. If your AI shortens it to a summary, the gate judges the summary.

Every response that runs opens with a status line such as Haakon Pre-Flight: Confidence 84/100 | Est. size: Medium (~5k-30k tokens, rough estimate) | Feasibility override: No, so anyone reviewing the conversation can see the score and whether the gate was overridden. With feasibility_off the line reads override: YES and the score is still shown. If the check itself is unavailable, the request runs and the line says so. The gate applies to the MCP tool only; the REST API is unchanged. Each call to Haakon counts as one request toward your hourly limit, whether it passes, is halted or is a standalone feasibility audit; the check itself is never counted a second time.

Cognitive Lenses

Pass these configuration flags into your IDE requests or MCP tool payloads to forcefully alter the LLM's reasoning framework.

Lenses combine freely in active_lenses, for example ["auto_ultimate", "socratic", "length_long"]. Auto Ultimate always applies the Universal base on non-coding questions, decides which of Strategy, Legal and Finance fit, then picks from the remaining lenses; it never picks a response-control or pre-flight lens, so add those yourself. Add one length lens (length_short, length_normal or length_long) to set the answer size; long has no limit. Extend widens the question to adjacent angles. Every request is also pre-checked by the Pre-Flight Feasibility Gate, which is on by default.