---
name: ask-cactus-help
description: Answer questions about Cactus from its published help, over MCP or llms.txt
---

# Ask Cactus help

## Over MCP

Connect to https://mcp.heycactus.ai/p/cactus/mcp (streamable HTTP, no sign-in). It answers with eight tools: `search_help`, `read_page`, `list_topics`, `read_timeline`, `ask_help`, `contact_team`, `search`, `fetch`.

## As text

https://cactus.heycactus.ai/llms.txt lists every published page. Each page's markdown is at https://cactus.heycactus.ai/<slug>/md, and https://cactus.heycactus.ai/llms-full.txt is every page in one file.

## As a git repository

`git clone https://cactus.heycactus.ai/memory.git` copies every published page into a folder in the Agent Memory Repo format: read `MEMORY.md` first, then follow its `[[path]]` links. Each publish reaches it within five minutes as one commit. It is read-only.

## In the browser

An agent running in a browser with WebMCP finds the same help as tools on every page of https://cactus.heycactus.ai, registered on `document.modelContext`: `ask_cactus`, `open_messenger`, `search_help`, `read_help_page`, `list_topics`, `read_timeline`. `open_messenger` is registered only where the Messenger is, on Cactus's own site. https://cactus.heycactus.ai/agents lists them and runs each one.

## Which tool

Use `ask_help` for a question that wants one answer drawn from several pages; it answers with the pages it cites. Use `search_help` to find the pages about a word or a feature, then `read_page` to read one in full.