Structured planning for AI-assisted development. PAPI gives Claude Code a persistent planning layer — every cycle builds on the last, so your project gets smarter over time instead of starting fresh every session.

Quick Start

Takes about 2 minutes.

Get your config — Sign up at PAPI and copy your .mcp.json snippet
Add to your project — paste it into .mcp.json in your project root
Restart Claude Code — it will detect the PAPI server automatically
In Claude Code: say run setup, then run plan

That's it. You're planning.

What You Get

Cycle planning — the AI planner reads your project history and recommends what to build next, with detailed BUILD HANDOFFs for each task
Build tracking — every task gets a build report: actual effort, surprises, discovered issues, architecture decisions
Strategy reviews — every 5 cycles, a deep analysis of velocity, estimation accuracy, and project direction
Compounding intelligence — cycle 20 is sharper than cycle 1 because it has 19 cycles of context to learn from

Before / After

Without PAPI	With PAPI
Start every session by re-explaining the project	Orient in one call — cycle state, next action, blocked tasks
Tasks grow in scope mid-build	BUILD HANDOFFs define scope boundary and acceptance criteria upfront
Architectural decisions get forgotten	Active Decisions persist across cycles with confidence levels
No way to know if you're going faster or slower	Estimation accuracy tracked every cycle

Telemetry

PAPI collects anonymous usage data (tool name, duration, project UUID — no code or task content). To opt out, add PAPI_TELEMETRY=off to your .mcp.json env block.

Hosted backend

PAPI is configured with PAPI_PROJECT_ID and PAPI_DATA_API_KEY — both are generated by the onboarding wizard at getpapi.ai and pasted into your .mcp.json. If those env vars aren't set, PAPI will fall back to local file storage (md mode) and emit a stderr warning that your cycles aren't visible on the dashboard. To get on the dashboard, sign up at getpapi.ai and use the config it gives you.