Run Proof where your agent already works.

Proof installs as an MCP server inside Claude Code, Claude Desktop, Cursor, and any other MCP-capable tool. It is free to install and use, and reading and writing your workspace runs on the model you already pay for. Your agent can frame a problem with Ideation, read the review record and leave its own notes for the person who signs, and pull the decision record out on demand. The app, the editor, and the agent all see the same workspace.

Set it up in a minute

Step 1Add Proof to your tool
Drop this into your tool's MCP config. For Claude Desktop, edit claude_desktop_config.json. For Claude Code and Cursor, add a .mcp.json at your workspace root.
```
{
  "mcpServers": {
    "proof": {
      "command": "npx",
      "args": ["-y", "@toremlabs/proof-mcp"]
    }
  }
}
```
Step 2Link your account
Run this once. It shows a short code, then open /device in Proof to confirm it. The tool is then linked to your account.
```
npx -y @toremlabs/proof-mcp@latest login
```

Ideation, Review, and the record, from your editor

Your agent reads the full workspace: every project, concept, and decision.
Ideation from the agent: frame a problem, explore approaches, compare them, and record the choice. The methods stay under the hood; you ask in plain language.
Every write to a concept is stamped with where it came from, so the record shows what the agent did and what you approved.
Pull the decision record out on demand: what was decided, on what evidence, when. Exportable, not locked in the app.
Reading and writing your workspace rides your host’s model, with nothing extra to buy. The idea-generation operators call a model of their own, using the provider key you saved.