PM Copilot

An AI-powered Product Management Operating System exposed as a Model Context Protocol server. It gives Product Managers, founders, and product leaders a set of tools, prompts, and reference resources for the workflows they do every day — PRDs, user stories, prioritization, competitor analysis, roadmaps, KPIs, meeting notes, and product strategy — directly from Claude Desktop, Claude Code, Cursor, Windsurf, or any other MCP-compatible client.

How generation works

PM-Copilot does not call out to an LLM API itself. Tools that produce narrative documents (PRDs, user stories, competitor analysis, roadmaps, KPIs, meeting summaries, strategy docs) use the MCP sampling capability to ask the connected client's own model to do the writing, using a PM-best-practices system prompt and a structured request. prioritize_features is fully deterministic: it computes RICE and ICE scores and MoSCoW buckets in code.

Tools

Tool	Purpose
generate_prd	Enterprise-grade Product Requirement Document
generate_user_stories	INVEST-compliant user stories with acceptance criteria and edge cases
prioritize_features	RICE / ICE / MoSCoW scoring and final prioritization
competitor_analysis	Feature comparison matrix, SWOT, and strategic recommendations
generate_roadmap	Quarterly roadmap with milestones, dependencies, risks
generate_kpis	North Star metric with leading/lagging indicators
analyze_meeting_notes	Summary, decisions, risks, action items, follow-ups
generate_strategy	Vision, mission, goals, market analysis, positioning, growth plan

Resources

• pm-copilot://templates/{prd,roadmap,user-story,strategy} — reusable document templates
• pm-copilot://frameworks — reference definitions for RICE, ICE, MoSCoW, Kano, JTBD, Opportunity Solution Tree, SWOT, Porter's Five Forces, and PESTLE

Prompts

create_prd, generate_user_stories_prompt, analyze_competitor, build_product_strategy, create_quarterly_roadmap — each wraps the corresponding tool with the right arguments.

Project structure

src/
  tools/       one module per MCP tool
  resources/   templates + framework library, registered as MCP resources
  prompts/     guided prompt wrappers around the tools
  schemas/     Zod input schemas and shared validation helpers
  utils/       structured logger and the sampling helper
  server.ts    builds and wires the McpServer
  cli.ts       stdio entrypoint (`pm-copilot-mcp` binary)
tests/         Jest unit tests

Development

npm install
npm run build   # compile TypeScript to dist/
npm test        # run the Jest test suite
npm run start   # run the compiled server over stdio

Docker

docker build -t pm-copilot-mcp .
docker run -i pm-copilot-mcp

The server communicates over stdio (the standard MCP transport for desktop clients), so run the container interactively (-i) rather than publishing a port.

Claude Desktop configuration

{
  "mcpServers": {
    "pm-copilot": {
      "command": "node",
      "args": ["/path/to/pm-copilot-mcp/dist/cli.js"]
    }
  }
}

Quality

• All tool inputs are validated with Zod; invalid input raises a descriptive error before any generation happens.
• Structured JSON logs are written to stderr so they never collide with MCP JSON-RPC traffic on stdout.
• Tool results return both human-readable text and structuredContent for programmatic clients.

Roadmap

Planned premium features include PM intelligence (usage/retention analytics), Jira/Notion/Linear integrations, and an AI product coach for interview and case-study practice — see the original product brief for the full commercialization plan.