Introduction
What PCR.dev is, why it exists, and how it works
PCR.dev is a human-first platform for Prompt & Code Reviews (PCRs), a practice we coined and validated with real engineers. If you are building software for real humans, humans should review the prompts powering that development at every step.
PCR.dev captures prompts you send to an AI coding assistant (Cursor, Claude Code, or any MCP-compatible tool) and organizes them as structured review artifacts alongside your code. Think of it as a commit history for the prompt layer: the reasoning and decisions behind each piece of AI-assisted code.
PCRs (Prompt & Code Reviews) is a practice coined and empirically validated with 20 software engineers, published at CHI 2026 (Best Paper Honorable Mention). PCR.dev is the infrastructure that makes the practice possible.
Key concepts
| Concept | Description |
|---|---|
| Project | A codebase you've registered with pcr init. Prompts from that directory are tagged to this project. |
| Session | A single conversation with an AI tool. Prompts within a session share a session_id. |
| Prompt | A single user message to the AI, along with the AI's response. |
| Source | The AI tool that generated the prompt: cursor, claude-code, mcp, or manual. |
| Capture method | How PCR captured the prompt: file-watcher (passive) or mcp (active tool call). |
Capture modes
PCR captures prompts in two ways:
- –File watcher (
pcr start): passively monitors the session transcript files that Cursor and Claude Code write to disk. Zero impact on your workflow. - –MCP server (
pcr mcp): an active integration where the AI tool calls PCR directly via the Model Context Protocol. Gives richer metadata.
Most users should start with the file watcher. It requires no changes to your existing AI tool setup and captures everything automatically. Your prompts will be ready for review when you are.