# ProductPilot MCP

An MCP server that helps digital product creators go from "what should I
build?" to a zipped, listing-ready product — end to end.

Built for creators shipping ebooks, prompt packs, wallpaper packs, Notion
templates, checklists, printable planners, cheat sheets, resume packs, and
study guides to Gumroad/Etsy.

## What it does

Ask Claude (or any MCP client connected to this server) things like:

- *"What digital product should I build today?"* → `find_trending_niches`, `score_opportunity`
- *"Is this niche worth pursuing?"* → `research_market`, `analyze_competition`, `analyze_reviews`
- *"Build me a complete product for this niche"* → `build_product_bundle` (the one-shot pipeline)
- *"Write the listing copy"* → `generate_listing`, `generate_keywords`, `generate_tags`
- *"What should this cost?"* → `generate_pricing_strategy`
- *"Prep this for Gumroad/Etsy"* → `prepare_gumroad_listing`, `prepare_etsy_listing`
- *"What sold well, and what should I build next?"* → `track_product_performance`, `recommend_next_product`

26 tools total — see `mcp_server/server.py` for the full list, or
`docs/ARCHITECTURE.md` for how they fit together.

## Quick start

```bash
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
python -m mcp_server.server
```

Runs entirely on deterministic mock market data out of the box — no API keys
required to try it. Full setup/deployment instructions: `docs/SETUP.md`,
`docs/DEPLOYMENT.md`.

## Using this on claude.ai (web)

Claude.ai's web app only connects to *remote* MCP servers (a public HTTPS
URL) — not local processes. This server supports that via
`PRODUCTPILOT_TRANSPORT=streamable-http`. Full walkthrough (deploy to Render
+ add as a claude.ai custom connector): **`docs/CLAUDE_AI_CONNECTOR.md`**.

## Status

Functionally complete: all 26 tools registered and runtime-tested end to end
(idea → outline → generation → SEO → zipped bundle → analytics tracking), 18
passing unit tests, Docker support.

**Known environment gotcha (already fixed here, documented for reference):**
the `mcp` PyPI package's `2.0.0` release renamed `FastMCP` to `MCPServer` and
dropped `mcp.server.fastmcp` entirely — `requirements.txt` pins
`mcp[cli]<2.0.0` to stay on the FastMCP-based API this server is built
against. See `docs/ARCHITECTURE.md` for the full explanation, including a
second subtle bug it's worth knowing about (`from __future__ import
annotations` breaking FastMCP's tool introspection).

## Project layout

```
mcp_server/    # all source (agents, market, research, scoring, generation,
               # seo, publishing, analytics, storage, config)
tests/         # pytest suite
docs/          # architecture, setup, deployment, roadmap
```

See `docs/ARCHITECTURE.md` for the full breakdown, `docs/ROADMAP.md` for
what's next.