# MCPqxbin

**MCP server for the QxBin Prompt Generator**

Probabilistic prompt engineering using **Binary Probability Matrices**.  
Room-temperature classical cubit logic that evolves and collapses optimal prompts.

By [Rupesh Malpani](https://x.com/rupeshmalpani) / [pikk.company](https://pikk.company)

---

## What it does

Treats prompt construction as a multi-dimensional probability landscape:

1. **Atom library** — Role / Reasoning / Structure / Quality / Constraint
2. **Binary Probability Matrix** — spatial grid of fractional probabilities
3. **Superposition** — `bias**n` + `(1-bias)**m` steers probability mass
4. **Measurement** — collapses to the highest-leverage prompt

Two modes:

| Tool | Mode | Description |
|------|------|-------------|
| `generate_optimal_prompt` | Edge | Single cubit, fast, seedable |
| `generate_ensemble_prompt` | Cloud | Parallel ensemble, picks global champion |
| `list_prompt_atoms` | — | Inspect the atom library |

---

## Quick Start (Local / Claude / Cursor)

```bash
# One-shot run (stdio — default)
uvx --from git+https://github.com/pikk-qxbin/MCPqxbin.git mcpqxbin

# Or install
pip install git+https://github.com/pikk-qxbin/MCPqxbin.git
mcpqxbin
```

### Claude Desktop / Cursor config (stdio)

```json
{
  "mcpServers": {
    "mcpqxbin": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/pikk-qxbin/MCPqxbin.git", "mcpqxbin"]
    }
  }
}
```

---

## Deploy to Railway (permanent public URL for Grok)

The repo is Railway-ready (Dockerfile + $PORT support).

### Steps

1. Go to **https://railway.app** and sign in (GitHub login works)
2. **New Project** → **Deploy from GitHub repo**
3. Select **`pikk-qxbin/MCPqxbin`** (or your fork)
4. Railway detects the Dockerfile and deploys
5. After deploy succeeds:
   - Open the service → **Settings** → **Networking**
   - Click **Generate Domain**
6. Your public MCP URL is:

```
https://YOUR-SERVICE.up.railway.app/mcp
```

### Add to Grok

1. Open → **https://grok.com/connectors**
2. **New Connector** → **Custom**
3. Paste: `https://YOUR-SERVICE.up.railway.app/mcp`
4. Save

Then just say:

> Use the QxBin prompt generator on this goal: …

---

## Local HTTP mode (testing / tunnel)

```bash
mcpqxbin --transport streamable-http --host 0.0.0.0 --port 8000

# then tunnel
ngrok http 8000
# or
cloudflared tunnel --url http://localhost:8000
```

---

## Example

**Tool call:** `generate_optimal_prompt`

```
goal: Write a sharp cold email to Mark Cuban about Pikkstops + QxBin
extra_context: Bootstrapped, under 180 words, first principles only
bias: 0.84
```

Returns a fully structured, high-signal prompt ready for any LLM.

---

## Core QxBin Math

- Fractional states: `bias**n` and `(1-bias)**m`
- Probability Matrix: 2D grid for multi-dimensional state
- Superposition blend + chain evolution
- Probabilistic measurement / collapse

Part of the [pikk-qxbin](https://github.com/pikk-qxbin) ecosystem.

---

## License

![License](https://img.shields.io/badge/License-Custom%20Dual%20Apache%2FMIT-blue)

This repository is released under the **official QxBin dual license** (Apache 2.0 **or** MIT) with commercial terms.

**Key terms:**
- **Free** for testing, experimentation, internal organizational use, and building your own software or improvements using your development resources.
- **51% revenue share** with the copyright holders applies when you create and sell a commercial tool, product, or API (whether for commercial customers or personal/end users) that incorporates or is substantially derived from QxBin logic.
- Enterprise-scale deployments and strategic partnerships are fully **negotiable** — reach out to [@rupeshmalpani](https://x.com/rupeshmalpani).

See the full [LICENSE](LICENSE) file for complete details.

Part of the pikk-qxbin vision: Democratizing advanced compute. Ship fast. Align incentives for long-term progress.