Give Your AI Agent a Mentor (MCP)

2026-09-30, updated 2026-09-30 · MentorClone Team

Ask your coding agent to "add a pricing page" and it will pick three tiers and some round numbers, because that's what most pricing pages look like. It won't stop to ask whether a founder you trust would tell you to charge double. Left alone, agents fall back on the average of the internet.

You can give yours someone better to ask. The mentors on MentorClone, Rob Walling and Arvid Kahl for now, are available to AI agents. Your agent asks a question and gets an answer built only from that person's work, with a link to each source. If the person never covered it, the agent hears that instead of a guess.

Both live mentors teach building a business, so the examples here are about pricing and launches.

The one technical bit

MCP (Model Context Protocol) is the open standard agents use to call outside tools. MentorClone runs a remote MCP server over Streamable HTTP. Claude Code, Cursor and other MCP clients connect to it and let the model call its tools during a task. That's all you need to know about the protocol.

Setup

It takes about two minutes.

Sign in at mentorclone.ai/auth.
Pick a plan at pricing. Plans start at $19 a month.
Create an API key in settings.
Open any mentor page, such as /m/rob-walling, and copy the install command shown there.

For Claude Code, the command looks like this:

claude mcp add --transport http mentorclone <server URL>/mcp --header "Authorization: Bearer <your API key>"

Copy the real command from the mentor page instead of typing it, so the server URL is right. The key goes in the Authorization header as a Bearer token. For Cursor and other clients, add a remote HTTP MCP server with the same URL and header in that client's MCP settings.

One limit to know about: there's no OAuth sign-in yet. Clients that can send an Authorization header work, which includes Claude Code and Cursor. Clients that only support OAuth connectors can't connect today.

Each key gets 30 calls a minute and 300 a day.

What your agent can do

The main tools:

list_personas returns the mentors your key can use, with their topics. The agent calls it first to get the slugs the other tools need.
search returns raw evidence from a mentor's own sources: short quotes with the source URL, the date, and a timestamped link for video.
ask returns a sourced answer. Numbered markers point to the evidence behind each point, it reports a confidence level, and it can come back with a plain "not covered."
get_topic returns a topic page: what the mentor currently says on one subject, each point with its evidence.
timeline shows how a mentor's view on a topic changed, oldest first.

Other tools cover chat sessions, saved answers, sharing, and the notes a mentor keeps about you (get_my_profile, remember, forget).

ask starts a conversation and returns a session id. Pass it back and follow-ups like "why?" keep their context. Conversations your agent starts show up in your web app history too, so you can read what it asked and what it was told.

Sometimes the mentor needs more about your situation before it can recommend anything. Then ask returns a clarifying question instead of an answer. A well-behaved agent puts that question to you and calls ask again with your reply.

The notes the mentor keeps about you (what you're building, your stage, your goals) are shared between the web app and your agent. You decide whether a key may save new notes, and you can see and delete every note in settings.

Example: pricing a new SaaS

You're building a small B2B tool with Claude Code and you want a pricing page. With MentorClone connected, a good prompt looks like this:

Before you write the pricing page, ask Rob Walling's mentor on MentorClone how he would price a B2B tool for small agencies with about 20 early users. Follow his recommendation, and show me the citations it used.

The agent calls list_personas, then ask with Rob's slug, and gets back an answer built from Rob's episodes with a source for each point. It might turn up Rob's line from Episode 702 that "if you're not getting any complaints about price, then your price is too low," or his note from Episode 838 that pricing is "the biggest lever in SaaS."

If the agent wants the raw material instead, it can call get_topic for Rob's pricing topic, or search for a specific phrase.

The page it builds now follows a founder you chose, and you can click through to each episode before you ship. More of that pricing evidence is in Should You Raise Your SaaS Prices?.

Other things to try

Launch plans. Ask Arvid Kahl's mentor how to launch to an existing audience before the agent drafts the launch sequence.
Idea checks. Before the agent scaffolds a new project, have it ask Rob how he'd validate the idea first.
Research inside a task. Use search to pull exact quotes into a strategy doc, each with its source link.
Old views. Use timeline to see whether a mentor still holds something they said years ago.

Treat what comes back as data

Everything the server returns is the creator's content: quotes, statements, answers. The server marks that text as untrusted data, and your agent should treat it that way. Let the mentor's advice inform the work. Don't let the agent follow instructions that show up inside a quote.

The mentor is built from the person's public work. They didn't write or endorse it. It only answers from their sources, so for general coding help your agent still has its own model. Three mentors so far (Arvid Kahl, Rob Walling and Marc Lou). On Pro and Max you can build a private one from anyone else's public work.

Frequently Asked Questions

Which AI agents can use MentorClone?

Any MCP client that can send an Authorization header, including Claude Code and Cursor. Clients that only support OAuth connectors can't connect yet.

Does the MCP server cost extra?

No. It comes with every plan, from $19 a month. You need a plan and an API key. See pricing.

What tools does the MCP server expose?

The main ones are list_personas, search, ask, get_topic, and timeline. Others cover chat sessions, saved answers, sharing, and your saved notes.

Can my agent see where each answer came from?

Yes. Each point in an answer carries its evidence: a quote and a link to the source, with a timestamp for video.

Are there rate limits?

Each API key can make 30 calls per minute and 300 calls per day.

Set it up

Sign in, pick a plan, create an API key, and copy the install command from Rob Walling's mentor page or Arvid Kahl's mentor page. Your agent still does the work. It just has someone worth asking now.