You're staring at a positioning doc, a launch plan, or a battlecard. You know the framework. You can't decide if you're applying it right.
PMM Sherpa is the senior PMM you wish you could DM at 11pm. It is a remote MCP server grounded in 38,000+ curated chunks from PMM books, AMAs, podcasts, and practitioner essays, with a calibrated advisory voice and four purpose-built tools. Connect it to whichever AI client you already use, and PMM judgment shows up in your normal flow, not in a new tab.
✨ The four tools
| Tool | When to use it | What it returns |
|---|
ask_sherpa | Strategic framing, "is this right?", "how should I think about X?", open-ended PMM judgment | Advisory dialogue grounded in the corpus, in Layer 4 voice |
draft_artifact | Generating one of 39 named PMM deliverables (positioning, messaging framework, ICP, persona, battlecard, landing page copy, launch plan, demo script, ad variants, comparison matrix, case study, analyst briefing, board deck, and more) | Structured deliverable with framework baked in |
get_feedback | Pressure-testing a draft, landing page, deck, or copy you already have | Critique against senior PMM heuristics: gaps + recommendations |
scope_pmm_research | Planning phase of a Deep Research run on a PMM-adjacent question | Research brief: angle, sub-questions, sources to weight, anti-patterns |
All four are grounded in the same corpus and speak the same voice. You orchestrate. Sherpa supplies the judgment.
Quick Start
Pick your client. Each setup takes a minute.
Claude.ai · Pro / Max / Team / Enterprise
- Claude.ai → Settings → Connectors → Add custom connector
- Name:
PMM Sherpa · URL: https://pmmsherpa.com/api/mcp · Auth: OAuth
- Sign in with the same Google account you use on pmmsherpa.com
- (Recommended) Upload the Claude.ai skill so Sherpa auto-triggers on PMM keywords with the right voice
- Ask: "Help me sharpen positioning for [your product]."
Claude Code · CLI + IDE extensions
claude mcp add pmm-sherpa --transport http https://pmmsherpa.com/api/mcp
OAuth opens in your browser on first call. Then install the skill so Sherpa auto-fires on PMM keywords:
git clone https://github.com/boommark/pmmsherpa-mcp.git ~/.pmm-sherpa
mkdir -p ~/.claude/skills
ln -s ~/.pmm-sherpa/skills/claude-code ~/.claude/skills/pmm-sherpa
In your next Claude Code session, prompt:
> See pmmsherpa.com. Help me create a battlecard against its top competitor.
ChatGPT · Plus / Pro / Team / Business — Developer Mode
- ChatGPT → Settings → Connectors → Advanced → toggle Developer Mode ON
- Settings → Connectors → Create:
- Name:
PMM Sherpa
- MCP Server URL:
https://pmmsherpa.com/api/mcp
- Authentication: OAuth, "I trust this application"
- Sign in with the same Google account as pmmsherpa.com
- Settings → Personalization → Custom Instructions → paste
skills/chatgpt/custom-instructions.md (~1,200 chars). This is the skill-equivalent for ChatGPT, since it does not yet support skills natively.
- In a new chat, toggle PMM Sherpa ON in the Tools panel. Ask: "Help me sharpen the positioning for [your product] against [competitor]."
Codex CLI, Gemini CLI, and Antigravity also work. See Multi-client install below.
🌐 Multi-client install
Sherpa works in every modern MCP-capable AI client. Pick yours.
1. Add the connector
Settings → Connectors → Add custom connector:
- Name:
PMM Sherpa
- MCP Server URL:
https://pmmsherpa.com/api/mcp
- Authentication: OAuth
Sign in with the same Google account as pmmsherpa.com. Approve.
2. Upload the skill (recommended)
Skills auto-trigger Sherpa on PMM keywords and enforce the Layer 4 voice across the conversation. Without one, you have to remember to toggle the connector and pick the right tool every time.
- Download
skills/claude-ai/pmm-sherpa-skill.zip from this repo
- Claude.ai → Settings → Capabilities → Skills → Upload skill
- Upload the zip → toggle ON
3. Try it
Help me build a battlecard against [competitor] for [your product].
The skill kicks in. PMM Sherpa is called automatically.
Full Claude.ai setup
1. Add the MCP server
claude mcp add pmm-sherpa --transport http https://pmmsherpa.com/api/mcp
On first call, Claude Code opens a browser for OAuth. Sign in with the same Google account as pmmsherpa.com.
2. Install the skill
git clone https://github.com/boommark/pmmsherpa-mcp.git ~/.pmm-sherpa
mkdir -p ~/.claude/skills
ln -s ~/.pmm-sherpa/skills/claude-code ~/.claude/skills/pmm-sherpa
The skill loads from .claude/skills/pmm-sherpa/SKILL.md on next session.
3. Try it
> /mcp
> See pmmsherpa.com. Help me create a battlecard against its top competitor.
Full Claude Code setup
1. Enable Developer Mode
Settings → Connectors → Advanced → toggle Developer Mode ON.
2. Add the connector
Settings → Connectors → Create:
- Name:
PMM Sherpa
- MCP Server URL:
https://pmmsherpa.com/api/mcp
- Authentication: OAuth
- Trust: "I trust this application"
Sign in with the same Google account as pmmsherpa.com.
3. Paste custom instructions
Settings → Personalization → Custom Instructions → paste the contents of skills/chatgpt/custom-instructions.md (~1,200 chars). This is the equivalent of the skill on ChatGPT, since ChatGPT does not yet support skills natively.
4. Try it
Help me sharpen the positioning for [your product] against [competitor].
Make sure the PMM Sherpa toggle is ON in the conversation's Tools panel.
Full ChatGPT setup
codex mcp add pmm-sherpa --transport http https://pmmsherpa.com/api/mcp
OAuth flow opens in browser on first call. Use the ChatGPT custom-instructions block in your ~/.codex/instructions.md to get skill-equivalent behavior.
Full Codex setup
Edit ~/.gemini/settings.json:
{
"mcpServers": {
"pmm-sherpa": {
"transport": "http",
"url": "https://pmmsherpa.com/api/mcp"
}
}
}
OAuth flow opens on first call. Paste the ChatGPT custom-instructions block into ~/.gemini/GEMINI.md for skill-equivalent voice.
Full Gemini CLI setup
Settings → MCP Servers → Add server:
- Name:
pmm-sherpa
- URL:
https://pmmsherpa.com/api/mcp
- Auth: OAuth
Full Antigravity setup
📦 Skills + custom instructions
Sherpa speaks a calibrated voice (Layer 4: discovery cadence, story-first, framework-named-mid, single-question close, never em-dashes). The skills enforce this voice and auto-route to the right tool.
Without the skill, Sherpa still works. You just have to remember to toggle the connector ON and pick the right tool. With the skill, it picks correctly on its own.
💡 Examples
Five sample workflows, end-to-end:
💰 Pricing
| Plan | Monthly | What you get |
|---|
| Free | $0 | 10 MCP credits / month. 2 credits per query. Try every tool. |
| Starter | $9.99 | 200 MCP credits / month. Unlimited use of pmmsherpa.com chat. Topup packs unlocked. |
Topup packs (one-time, never expire):
| Pack | Credits | $/credit |
|---|
| $5 | 50 | $0.10 |
| $10 | 125 | $0.08 |
| $15 | 200 | $0.075 (best value) |
See full pricing · Purchase docs
📚 The corpus
38,000+ chunks, hand-curated across the highest-signal PMM material in the field:
- 34 authoritative books on positioning, messaging, pricing, GTM, and launch
- 1,100+ practitioner conversations: podcast episodes and senior-PMM AMAs
- 1,600+ practitioner essays: long-form blog posts and operator write-ups
The corpus is curated, not crawled. Every source is hand-picked for substance density. Want something added? Open a corpus request.
🎙️ Voice & cadence
Sherpa speaks in Layer 4 voice. Every response:
- Opens with your situation, not preamble. No "Great question."
- Tells the story before naming the framework.
- Varies paragraph length. Long build. Short punch. Question. Space.
- Closes with one question you take away. Never a menu of options.
- Never uses em-dashes. Period, comma, colon, or parentheses instead.
- References principles. Never names the authors, books, or podcasts in the corpus.
- Doesn't hedge ("it depends", "could be") when the corpus has a clear position.
Same voice senior PMMs use in private Slack DMs. It transfers if you install the skill.
❓ FAQ
Each MCP tool call is 2 credits, flat. Free users get 10 monthly credits (resets on your anniversary date). Starter users get 200 monthly credits + can buy topup packs that never expire. Founders are unmetered.
No. The corpus and the prompt-engineering live on Sherpa's servers. The MCP transport is the supported integration point.
Your prompts and Sherpa's responses are logged for observability (Langfuse traces tagged surface:mcp). We do not train models on your data. We do not sell or share with third parties.
MCP connectors are gated to Pro/Max/Team/Enterprise on Claude.ai. Same on ChatGPT (Plus/Pro/Team/Business). Free tiers across major LLM providers do not support custom MCP connectors yet.
Yes. New books, AMAs, and high-signal essays are added every month. Open a corpus request issue to flag a gap.
The voice is calibrated to teach principles, not parade sources. Naming authors creates false authority and breaks reading flow. Sherpa references the principle. The principle holds whether or not the source is named.
🤝 Contributing
This repo is the public docs and skill mirror. The Sherpa server itself is closed-source. We welcome:
- Bug reports and feature requests for the MCP server, skills, or docs (open an issue)
- Corpus requests, books, AMAs, podcasts you want added (open a corpus request)
- Doc PRs for the skill files, examples, or README
See CONTRIBUTING.md for the full guide.
📜 License
MIT for the docs, skills, and examples in this repo. The Sherpa server, corpus, and proprietary prompts are not open-sourced.
🙏 Acknowledgements
- The PMM community whose books, AMAs, podcasts, and essays form the corpus
- The MCP spec authors at Anthropic for designing the protocol that makes this possible
- The Vercel AI SDK team for the streaming + tool-calling primitives
- The Supabase team for pgvector
- Every PMM who beta-tested Sherpa and pushed back on bad outputs
