@unfakable/mcp

Document verification MCP server for AI agents.

@unfakable/mcp is a Model Context Protocol server that gives AI agents (Claude Code, Claude Desktop, Cursor, ChatGPT, Gemini CLI, any MCP host) a primitive answer to one question: is this file the record it claims to be?

It does that by computing a SHA-256 fingerprint of the file locally — the file never leaves the device — and checking whether the fingerprint is anchored to a Bitcoin block via OpenTimestamps on the Unfakable verifier index. If the file has been locked, you get back the anchor block, the locked-at timestamp, the counterparty notification status, and a permanent verifier URL.

Zero file storage. Independent verification. Built for the era where AI agents read, summarize, sign, route, and trigger workflows from documents that increasingly cannot be trusted on their face.

Five tools

| Tool | What it does | | --- | --- | | verify_file(path) | Hash a local file with SHA-256 and check whether Unfakable has a Bitcoin-anchored proof for that fingerprint. The file never uploads. | | verify_url(url) | Fetch a public URL, compute its fingerprint, and check the index. Useful for verifying remote PDFs, wire instructions sent over email links, etc. | | verify_hash(hash) | Direct lookup of a known SHA-256. Use when you already have the hash from another system. | | lock_file(path, ...) | Create a new Bitcoin-anchored lock for a local file. Requires UNFAKABLE_API_KEY. Free key at https://unfakable.ai/lock. | | get_evidence_pack(handle) | Retrieve Evidence Pack metadata (19-file manifest, FRE 902(14) cert info, chain-of-custody) for an existing lock. |

Install

Claude Code

claude mcp add unfakable

Or manually, in your Claude Code MCP config:

{
  "mcpServers": {
    "unfakable": {
      "command": "npx",
      "args": ["-y", "@unfakable/mcp"],
      "env": {
        "UNFAKABLE_API_KEY": "YOUR_KEY_HERE"
      }
    }
  }
}

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "unfakable": {
      "command": "npx",
      "args": ["-y", "@unfakable/mcp"],
      "env": {
        "UNFAKABLE_API_KEY": "YOUR_KEY_HERE"
      }
    }
  }
}

Restart Claude Desktop. The five tools appear in the MCP servers list.

Cursor

In Cursor → Settings → MCP, add:

{
  "mcpServers": {
    "unfakable": {
      "command": "npx",
      "args": ["-y", "@unfakable/mcp"],
      "env": { "UNFAKABLE_API_KEY": "YOUR_KEY_HERE" }
    }
  }
}

Gemini CLI

Add to ~/.gemini/extensions.json:

{
  "mcpServers": {
    "unfakable": {
      "command": "npx",
      "args": ["-y", "@unfakable/mcp"],
      "env": { "UNFAKABLE_API_KEY": "YOUR_KEY_HERE" }
    }
  }
}

Standalone

npm install -g @unfakable/mcp
unfakable-mcp

Speaks JSON-RPC over stdio. Pipe it into any MCP-compatible host.

Authentication

• verify_file, verify_url, verify_hash, get_evidence_pack — no key required. All four are read-only against the public verifier index.
• lock_file — requires UNFAKABLE_API_KEY. Free key at https://unfakable.ai/lock. The key authorizes lock creation against your Unfakable account, applies your dispute window settings, and emails counterparty notifications under your domain.

Why this exists

For thirty years, document proof has lived inside the vendor that holds the file — the inbox provider, the signature platform, the document portal. Breach the vendor, subpoena the vendor, retire the product, and the proof goes with it.

AI agents now read, summarize, sign, and act on documents at scale. Before an agent acts, it needs to verify the file is the record it claims to be. That primitive does not exist inside any vendor today.

@unfakable/mcp is that primitive. It lets any agent — Claude, Gemini, ChatGPT, Cursor — call a single tool and get back: locked, locked when, anchored to which Bitcoin block, verifier URL, counterparty status. The proof survives if Unfakable disappears because the anchor lives on a chain Unfakable does not control.

Links

• Site: https://unfakable.ai
• MCP docs: https://unfakable.ai/mcp
• Repo: https://github.com/unfakable/mcp
• Issues: https://github.com/unfakable/mcp/issues
• Smithery: https://smithery.ai/server/@unfakable/mcp

License

MIT — see LICENSE.