MCP

Use AI PM Lab inside your AI assistant

Add AI PM Lab to Claude, ChatGPT, Cursor or VS Code, and your assistant can draw on the lessons while you work: look up a concept, price a feature, review a spec or design an eval suite. Every answer links back here, where you can try the idea with real models.

Server URL
https://aipmlab.io/api/mcp

Try asking

  • “Review this AI feature spec with AI PM Lab and tell me what's missing.”
  • “What would 50,000 requests a day cost on gpt-4.1-mini with a 3,000-token prompt?”
  • “Design an eval suite for our support bot that answers from the help docs.”
  • “Replay an AI PM Lab prompt-injection run where only the input guardrail is on.”

What it can do

Look things up

  • Find the lesson on a topic, with its key idea and what to try
  • Define 60+ AI concepts in plain words
  • List the challenges

Run the numbers

  • Count the tokens in a prompt
  • Estimate what a model costs per request, day and month, with live prices

See real results

  • Replay recorded real experiments: did the prompt injection leak? did retrieval find the answer?
  • Browse the runs recorded on each 3D page, from agent loops to evals, and watch any of them in 3D

Do the work

  • Review an AI feature spec against what the lessons teach
  • Prepare questions for a planning meeting with engineers
  • Design an eval suite: test cases, graders, rubric and pass bars

Add it

Claude (web and desktop)

Settings → Connectors → Add custom connector. Name it AI PM Lab and paste the server URL. Keep No sign-in, which Claude detects, and click Add: the warning that anyone with the URL can use it is expected, since the server is free and open and its tools only read. On Team and Enterprise plans, an owner adds it first.

ChatGPT

Settings → Apps & Connectors → Advanced settings → turn on Developer mode. Then Create, paste the server URL and choose No authentication.

Claude Code

claude mcp add --transport http ai-pm-lab https://aipmlab.io/api/mcp

Cursor

Settings → MCP → Add new MCP server, or add this to your mcp.json:

"ai-pm-lab": { "url": "https://aipmlab.io/api/mcp" }

VS Code

code --add-mcp '{"name":"ai-pm-lab","type":"http","url":"https://aipmlab.io/api/mcp"}'

Any other app that supports remote MCP servers over HTTP works the same way: give it the server URL.

Tools

Your assistant picks these on its own when you ask; you never need their names. They're listed here so you can see exactly what it can do.

Look things up

  • search_lessons

    Finds the lessons on a topic, best match first, with each one's key idea.

    “Which AI PM Lab lesson covers hallucinations?”

  • get_lesson

    A lesson's key idea, concepts and things to try, with links to the lesson, its 3D tour and its challenge.

    “Summarise the RAG lesson and what I should try in it.”

  • define

    A plain-language definition of 60+ AI concepts, with the lesson that teaches each.

    “What's LLM-as-judge, in plain words?”

  • list_challenges

    The challenges and leaderboards, and what each one asks you to do.

    “What challenges can I try on AI PM Lab?”

Run the numbers

  • count_tokens

    Counts the tokens in a text. Nothing you send is kept.

    “How many tokens is this system prompt?”

  • estimate_cost

    What a model costs per request, day and month at your volume, from live prices, with prompt caching.

    “What would 50,000 requests a day cost on gpt-4.1-mini with a 3,000-token prompt?”

See real results

  • list_experiments

    The real runs recorded on a 3D page (prompt injection, RAG, evals, agent loop, next token, embeddings).

    “What prompt-injection experiments has AI PM Lab recorded?”

  • replay_experiment

    What happened in one recorded run, and why, with a link to watch it in 3D.

    “Replay a prompt-injection run where only the input guardrail is on.”

Do the work

  • review_ai_specalso a prompt

    Checks your spec against what the lessons teach, from evals to injection defences, and links each gap to its lesson.

    “Review this AI feature spec and tell me what's missing.”

  • prep_planning_meetingalso a prompt

    The questions to ask your engineers about a planned AI feature, and what good answers sound like.

    “Prep me for a planning meeting about our AI search feature.”

  • design_eval_suitealso a prompt

    Builds an eval suite for your feature: test cases, graders, a judge rubric, pass bars and a JSON test set.

    “Design an eval suite for our support bot that answers from the help docs.”

In Claude, the three marked “also a prompt” appear as ready-made prompts you can pick, as well as tools the assistant calls itself.

Privacy

The connector answers from AI PM Lab's own lessons, prices and recorded runs; it never sends what you ask to another AI model. Your assistant does the thinking. Nothing you send is stored: text you count tokens for is counted and forgotten. The site only counts how often each tool is used, with no text and nothing about you. See the privacy notice.