Fractal Process Intelligence — Research Alpha

Fractal measures whether additional AI actually created additional intelligence.

Fractal is not trying to make AI smarter. It observes research process and produces an X-Ray of what a polished final answer may hide.

Primary question: Did additional AI actually create additional independent intelligence?

This repository is the public distribution surface for the Fractal Research MCP (recorder/observer). The live remote endpoint is already hosted; this repo documents how to connect and give feedback.

Research Alpha: This is an active research experiment, not a validated commercial product.

What it does

Fractal Process Intelligence MCP is recorder/observer only.

Compatible AI clients can record observable research-process artifacts:

• participating agents

• meaningful sources

• claims

• uncertainty

• disagreement

• tool usage

Then finish_run compiles a process X-Ray (hosted report URL).

Public product page: fractalinfinity.io/research-lab.html
Sample X-Ray: fractalinfinity.io/report/sample.html

Related MCP server: MCP Execution Ledger

What it does NOT do

• Call AI models (no Anthropic / OpenAI / Gemini inference)

• Buy model inference on your behalf

• Require provider API keys

• Inspect hidden chain-of-thought

• Automatically spawn agents

• Estimate Claude/OpenAI billing or token dollar cost

Health check advertises "inference":"none".

Public MCP endpoint

https://mcp.fractalinfinity.io/mcp

Health: https://mcp.fractalinfinity.io/health

Current recorder tools

Claude Desktop (Windows) alpha setup

1. Claude Desktop → Settings → Developer → Edit Config

2. Merge the MCP entry below (do not wipe other config keys).

3. Save.

4. Fully quit Claude (system tray → Quit).

5. Reopen Claude.

6. Settings → Developer should show fractal-process-intelligence running.

{
  "mcpServers": {
    "fractal-process-intelligence": {
      "command": "C:\\Program Files\\nodejs\\npx.cmd",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.fractalinfinity.io/mcp"
      ]
    }
  }
}

See also: examples/claude-desktop.mcp.fragment.json

Microsoft Store builds of Claude may store config under a package LocalCache path; prefer Edit Config so Claude opens the file it actually reads. Use the full npx.cmd path on Windows.

Example research instruction

Paste into a new research conversation after connecting:

Use Fractal Process Intelligence to record this research run.

Record the participating agents, meaningful sources, important claims,
uncertainty, disagreement and significant tool usage.

Finish the run and return the Fractal X-Ray URL.

Do not interpret the X-Ray for me before I open it.

Then ask your normal research question. Read the AI’s final answer first. Open the X-Ray second.

Privacy / security boundary

Fractal records only artifacts explicitly sent through its MCP tools.

It does not automatically record every Claude (or other) conversation. For normal chats, do not start a Fractal run / do not invoke its tools.

Feedback (the research question)

Open a GitHub Issue or reply in discussions.

Please answer:

1. Did the X-Ray reveal something consequential you would not have known from the final AI answer alone?

2. Would you want this attached to your normal research or agent workflow?

Optional: what was confusing or misleading?

Links

License

MIT — see LICENSE.