summon_legend and get_legend_context hand back the persona and stop there; nothing in the tool list generates the reply, so what you read is your own model wearing the voice. The roster is the fixed set named in list_legends, and the framing is unmistakably proactive — suggest is written to be called on questions about business, startups, investing, crypto, AI, leadership and life advice, and returns an action for the client to execute rather than an answer to read.
A persona library: the tools return a named founder's or investor's identity, voice, principles and thinking frameworks as context, and your model answers in that character.
- A fixed roster you can browse: list_legends groups it into Tech Titans (Elon Musk, Steve Jobs, Jeff Bezos, Jensen Huang), Investors (Warren Buffett, Charlie Munger, Peter Thiel, Marc Andreessen), Startup Sages (Paul Graham, Sam Altman, Naval Ravikant, Reid Hoffman) and Crypto Builders (CZ, Anatoly Yakovenko, Mert Mumtaz, Michael Heinrich).
- search_legends finds them by name, description, expertise or tags — 'first principles' resolves to Elon Musk, 'investor' to Buffett, Munger and Thiel.
- summon_legend returns one legend's full persona context so the reply comes back in that voice, and get_legend_context returns the long form: identity, voice and style, core principles, thinking frameworks and anti-patterns.
- party_mode puts one question to several legends at once and returns each of them answering in their own voice.
- auto_match reads a question, matches legends by expertise and returns them with key insights; get_legend_insight returns a single wisdom snippet on a topic; suggest returns a ready-to-execute primary_action naming the tool to call next.
No account and no key. The tools return persona text for the model to answer with, so the quality of the reply depends on the model in your client.
One command — npx legends-mcp