Make your product answerable to compatible AI clients
Your customers ask ChatGPT, Claude, and Perplexity before they ask you — and those agents may research on their behalf. HelpShelf publishes a live MCP server and llms.txt surface so compatible clients can query your sources through four read-only tools.
Can AI agents use HelpShelf?
Yes. HelpShelf gives every site a live MCP server and an llms.txt context bundle, so AI agents and clients can search published product context. Supported answers include citations, and the public endpoint cannot create leads, tickets, messages, or handoffs. HelpShelf generates the surface, with a Free plan to start.
Your customers ask AI before they ask you.
Search is becoming a conversation with an agent — and that agent decides what your product is.
- An AI agent answers for you — from a stale blog post, a competitor, or a hallucination.
- The agent hits an undocumented question and needs a separate human-facing contact path.
- You have a help center for humans, but nothing structured for the machines reading on their behalf.
- You cannot see what agents are asking, or where they get your product wrong.
From content to agent-ready in three steps
No SDK to integrate. No server to run.
Publish your agent surface
HelpShelf turns your content into an llms.txt bundle and a hosted MCP server automatically — no engineering required. Curated and published content only.
Agents query your product
Compatible MCP clients can search published docs, ask questions, and pull product context. Supported answers include citations; blocked or unsupported replies may not.
Agents inspect the evidence
Clients can follow returned citations or fetch an article in full. If the sources do not support an answer, they must use a separate human-facing contact path.
Everything an AI agent needs to get you right
Readable, cited when supported, and read-only at the public endpoint.
A live MCP server
Every site gets a hosted MCP endpoint with search_docs, ask, get_article, and get_product_context so compatible MCP clients can query published product knowledge.
llms.txt, generated for you
HelpShelf publishes llms.txt and a full llms-full.txt context bundle from your verified content, so agents read an accurate, curated version of your product.
Cited answers, not guesses
Agents get supported answers grounded in your published sources with citations. When the evidence is insufficient, the endpoint reports that limit instead of presenting an unverified feature claim.
Read-only by design
The public endpoint cannot create a lead, ticket, message, or handoff. External clients can inspect published knowledge without gaining an action surface.
One source of truth
The same published knowledge powers your widget, help center, llms.txt, and MCP endpoint; update the source set once for those surfaces.
A machine-readable source
Structured llms.txt and MCP surfaces give compatible tools a source they can inspect. They do not guarantee discovery, ranking, recommendation, or citation by an outside model.
Usable by compatible clients
One MCP + llms.txt surface that compatible tools can inspect when configured or linked
Frequently asked questions
Yes. Every site gets a live MCP server and an llms.txt context bundle. Compatible clients can search published product knowledge, and supported answers include citations. The four public MCP tools are read-only.
MCP (Model Context Protocol) lets compatible AI clients call tools and pull external context. HelpShelf exposes four read-only tools: search_docs, ask, get_article, and get_product_context.
Any tool that supports MCP or can read a published llms.txt — including ChatGPT, Claude, Perplexity, and AI code editors like Cursor and Windsurf. HelpShelf also adds Ask-with-my-AI buttons that deep-link your help content into those assistants.
No. Point HelpShelf at your content and it generates the llms.txt bundle and hosts the MCP endpoint for you. There is no SDK to integrate and no server to run — the agent surface goes live alongside your widget.
A chatbot answers humans on your site. The agent surface gives compatible outside clients published sources they can query. Supported MCP answers include citations, and the public endpoint remains read-only rather than creating leads, tickets, messages, or handoffs.
It creates structured, current, machine-readable surfaces that compatible tools can inspect. HelpShelf does not claim that publishing llms.txt or MCP forces an outside model to discover, rank, recommend, or cite your product.
Support for the humans, too
The same published source set powers supported widget answers, your hosted help center, llms.txt, and four read-only MCP tools.