Home / How to Use InitRepo: A Complete Guide to Getting the Best Results

How to Use InitRepo: A Complete Guide to Getting the Best Results

Step-by-step guide to filling out the InitRepo questionnaire for maximum document quality, then using your planning suite with Claude Code, Codex, OpenCode, and the InitRepo MCP server.

The one thing to know before you start

InitRepo's eight documents are only as good as the questionnaire answers you give it. A generic brief produces generic requirements. A specific, opinionated brief produces planning documents that an AI coding agent can implement directly — no clarification needed. The tips below are the difference between documents you reference once and documents you use every day.

How to fill out the questionnaire for best results

The questionnaire walks you through product context, audience, technical constraints, and scope. Here is what separates high-quality output from generic boilerplate.

1

Name the exact problem, not the product category

"A task manager" is thin. "A task manager for freelance designers who bill by project and need to track time against client budgets" gives the AI a real problem to write requirements around.

2

Specify your stack — even if partial

If you know you're using Next.js and Postgres, say so. If you're undecided, say "undecided" rather than leaving it blank. The architecture doc generates against your constraints.

3

Describe your user in one sentence

One clear persona beats a laundry list. "Solo founders validating MVPs before hiring" tells the AI what to optimise for. It will infer pain points, goals, and vocabulary from that anchor.

4

List what you've already decided

Auth provider, payment processor, hosting, design system — decisions already made are constraints the AI should honour, not revisit. Naming them prevents contradictory recommendations.

5

Give the monetisation model

Subscription, usage-based, one-time, or freemium changes the PRD significantly. A SaaS product and a one-time-purchase tool have different acceptance criteria for the same feature.

6

State your launch scope explicitly

"MVP for first 50 beta users" and "full commercial launch" generate very different roadmaps. The narrower and more honest your scope, the more usable the phased plan.

The "Improve" button

After filling out the questionnaire you'll see an ✨ Improve option on each field. Use it — it rewrites your answers to be more precise and detailed without changing your intent. Run it on any field that felt vague when you wrote it.

Your eight planning deliverables

One questionnaire run generates eight cross-referenced documents. Each one is a structured planning artefact designed to be read by both humans and AI coding agents.

Execution Blueprint

The master narrative your agent reads first — product vision, priorities, and decision rationale in one place.

Business Analysis

Market positioning, problem definition, user personas, and success metrics.

PRD

Full product requirements: features, user needs, and acceptance criteria in structured form.

UX/UI Spec

Design direction, component library choice, and screen-by-screen UI block specifications.

Technical Architecture

Stack decisions, data model, API design, and component structure — all decided before you write a line of code.

User Stories

Every feature broken into implementable stories with P#S# coordinate addresses and acceptance criteria.

Roadmap

Phased implementation plan with sequenced milestones and priorities.

Context Index

A cross-document index keyed by P#S# coordinates so agents can retrieve exactly the right context for any task.

Documents use a shared P#S# coordinate system (Phase number, Step number). Every user story, every architecture decision, and every roadmap milestone references the same coordinates — so an AI agent can retrieve exactly the right context for any implementation task.

Using your documents with AI coding tools

Download your documents as a ZIP, extract them to a docs/ folder in your repository root, and commit them. From there, every AI coding tool can read them — here is how each major tool works best.

Claude Code / Claude

Claude Code automatically indexes files in your repository. Add your docs/ folder and Claude can reference any document by name in its context window.

  1. 1Extract the ZIP into docs/ at your repo root and commit.
  2. 2Create or update CLAUDE.md to include: See docs/ for the project planning suite — read relevant documents before implementing any feature.
  3. 3Reference stories by coordinate in your prompts: Implement P1S2 from docs/user-stories.md.
  4. 4Use the InitRepo MCP server to skip manual file references entirely — see below.
GitHub Copilot / Codex

GitHub Copilot indexes your workspace files for completions, and Codex reads repository content before generating code. Place your documents where both tools can reach them.

  1. 1Extract the ZIP into docs/ and commit to the repo — Copilot's workspace index picks these up automatically.
  2. 2Open the relevant document in a VS Code tab before starting a Copilot Chat session — the open tab is included in context.
  3. 3For Codex, attach the architecture and user stories files when configuring your task — Codex reads attached files before writing code.
OpenCode

OpenCode is a terminal-first AI coding agent. It operates on your repository files and accepts context documents the same way Claude Code does.

  1. 1Extract the ZIP into docs/ and commit so OpenCode can read the files in context.
  2. 2Add an AGENTS.md file at the repo root instructing OpenCode to read docs/ before implementing features.
  3. 3Reference specific documents in your task prompt: "Implement the feature described in P1S3 of docs/user-stories.md, following the stack in docs/architecture.md."

The InitRepo MCP server — zero-copy context

Instead of manually adding documents to each agent session, the InitRepo MCP server exposes your project's planning suite as live resources that any MCP-compatible agent can read on demand. Configure it once; every session from that point reads your current documents automatically.

Always current

The MCP server reads from your live InitRepo project. If you regenerate a document, your agent picks up the new version on its next request — no re-downloading, no stale copies.

Coordinate-indexed retrieval

Agents can query the Context Index by P#S# coordinate and retrieve exactly the right context for a given task, rather than loading the entire document suite.

Works with any MCP client

Claude Code, Cursor, Cline, Continue.dev, and any other tool that supports the Model Context Protocol can connect to your InitRepo project through the same server.

Connect in three steps

  1. 1Install the bridge: npm install -g initrepo-mcp
  2. 2Copy your Project ID and API key from the InitRepo dashboard.
  3. 3Add the server to your agent's MCP config (.claude/settings.json for Claude Code, MCP settings for Cursor or Cline) — see the full setup guide.

What changes when you use InitRepo

Spend 30 minutes briefing your agent at the start of every session

Documents are the persistent brief — your agent reads them and starts implementing

Agent invents your data model, stack decisions, and API shapes

Every decision is in the architecture spec — the agent implements against your design

Features drift from requirements as the project grows

Every feature maps to a user story with acceptance criteria the agent can verify

Architecture decisions made implicitly in code

Architecture decided upfront, implemented consistently across the full codebase

Ready to build with a proper plan?

A Max trial gives you all eight planning deliverables — Execution Blueprint, Business Analysis, PRD, UX/UI Spec, Architecture, User Stories, Roadmap, and Context Index — AI-written from one questionnaire. No card games: it's 3 days free, then see pricing to continue.

3-day free trial · card required to start · cancel any time

See the AI coding tool guides for tool-specific setup instructions, or the MCP server page for the full connection walkthrough.

Related reading