Connect ChatGPT

Add PMM Sherpa to ChatGPT via Developer Mode.

ChatGPT supports remote MCP in two places: Deep Research / Company Knowledge (GA) and Developer Mode (beta, full read+write tool calls in regular chat). For PMM Sherpa, use Developer Mode. It is the only path that exposes all four tools.

Note: ChatGPT renamed "Connectors" to Apps in late 2025.

Available on Plus, Pro, Business, and Edu. Not available on Free.

  1. Settings → Apps → Advanced settings → Developer mode → enable.
  2. Settings → Apps → Add custom app (or "Add MCP server" inside Developer Mode).
  3. Enter:
    • URL: https://pmmsherpa.com/api/mcp
    • Auth: choose OAuth.
  4. Sign in with your PMM Sherpa account in the popup.
  5. In a new chat, open the apps/tools menu and toggle PMM Sherpa on.

ChatGPT does not yet support skills natively. The closest equivalent is Custom Instructions, a free-text field that's prepended to every conversation. The block below replicates skill behavior: tool routing + voice rules. ~1,200 characters, fits inside ChatGPT's field limit.

Settings → Personalization → Custom Instructions → paste the block below into "Anything else ChatGPT should know about you?" or "What traits should ChatGPT have?"

PMM Sherpa connector: call it for positioning, messaging, GTM, launch,
ICP, persona, battlecard, pricing, narrative, win/loss, analyst.
- ask_sherpa: advisory dialogue, framing, "is this right"
- draft_artifact: when I name a deliverable in the 39-template catalog
  (positioning, messaging framework, ICP, battlecard, landing page,
  launch plan, demo script, etc.). Pass artifact_type.
- get_feedback: pressure-test my existing work
- scope_pmm_research: only at Deep Research planning phase

Voice (everything in a Sherpa thread):
- No em-dashes. Use period, comma, colon, or parentheses.
- Open with my situation. No "Great question."
- Story before framework name.
- Vary paragraph length. Long build. Short punch.
- Close with one question. Never an option menu.
- Don't narrate Sherpa calls. Present the response as the response.
- Don't name authors, books, podcasts, or companies from the corpus.
  Reference principles instead.
- Don't hedge when the corpus has a clear position.

The same block works on Codex CLI (~/.codex/instructions.md), Gemini CLI (~/.gemini/GEMINI.md), and Antigravity (workspace-level INSTRUCTIONS.md).

Verify#

Open a new conversation. Make sure the PMM Sherpa toggle is ON in the Tools panel. Send:

Help me sharpen the positioning for [your product] against [competitor].

You should see ChatGPT call ask_sherpa (or draft_artifact). Response in Layer 4 voice: situation-first, no em-dashes, single closing question.

<img src="/pmmsherpa_mcp_chatgpt.png" alt="PMM Sherpa connector running in ChatGPT" />

Auth#

Developer Mode uses in-browser OAuth. The legacy Connector mode also accepts a pasted bearer token if your workspace blocks OAuth (rare).

Important Deep Research caveat#

ChatGPT's Deep Research surface only accepts MCP tools that match a search / fetch shape. That means scope_pmm_research, draft_artifact, and get_feedback will not appear inside Deep Research. They only work in regular chat with Developer Mode on.

If you specifically want PMM Sherpa during a ChatGPT Deep Research run, plan up front in regular chat (use scope_pmm_research and ask_sherpa to brief), then run Deep Research, then return to regular chat for get_feedback on the synthesis.

Troubleshooting#

  • Developer Mode toggle missing: confirm your plan (Plus, Pro, Business, or Edu). Free does not see Developer Mode.
  • OAuth popup blocked: allow pop-ups for chatgpt.com and pmmsherpa.com, then re-run any tool call.
  • App not appearing in chat: open the apps menu inside the conversation and toggle PMM Sherpa on. Each chat is opt-in.
  • Want staging: use https://staging.pmmsherpa.com/api/mcp and a staging account.

Next steps#

Back to top

PreviousConnect Claude CodeNext Connect Codex