Connect Claude, or an agent you build, to your knowledge base. Ask what your customers searched for and could not find, and get the missing articles drafted for your review. 29 tools, inside the permissions you approve.
Included in Growth and Catalyst, and in their 14-day trials. No credit card needed.
Claude on the web or desktop, Claude Code, or an agent you build.
https://mcp.helpcenter.io
Paste it into Claude as a custom connector.
Works with Claude on the web, Claude Desktop and Claude Code, and with any MCP client that supports remote servers with OAuth.
Every search that comes back empty is a customer telling you what to write next. HelpCenter.io keeps count. Your agent reads that list ranked by demand, checks what you already have, and drafts the articles worth adding.
New articles can be saved as drafts, so going live stays your call. The next customer who asks gets an answer instead of a dead end.
Search data is aggregated daily: your agent sees questions and demand, not a trail of individual visitors.
You: What did our customers search for last month and not find?
Tool call:
helpcenter_get_search_queries
reads analytics
Assistant: Three searches came up empty most often last month:
Tool call:
helpcenter_search_articles
×3, reads content
Assistant: Pausing is already covered in “Hold your plan”, it just never says “pause”. I can add the word there instead of writing a duplicate. The other two have no article yet. Shall I draft them?
You: Yes, draft both.
Tool call:
helpcenter_create_article
×2, saved as drafts
Assistant: Two drafts are waiting for your review. Nothing goes live until you publish them.
MCP, the Model Context Protocol, is an open standard that lets an AI assistant use tools outside the chat. These are the tools HelpCenter.io gives it, and every one of them runs inside the permissions you approve.
6 tools
Search, list and read every article, create new ones and update the ones you have. Import up to 100 at a time, and give each a stable external ID so a re-run updates what it made instead of duplicating it.
3 tools
A traffic and ratings summary, content performance article by article, and the searches visitors ran that found nothing, ranked by demand.
6 tools
Stage an edit to a live article privately, then publish it or discard it. List an article's earlier versions and restore one.
6 tools
List, create, update and delete categories, and reorder them and the articles inside them. A category that still holds content is not deleted until you choose where that content goes.
4 tools
List reader comments across your help center or on one article, moderate them, and reply as yourself.
3 tools
Read, answer and resolve the private notes your team leaves on articles. A permission of its own, never bundled into content access.
1 tool
List the help centers a connection can reach. Connect several and one conversation can work across all of them.
In Claude on the web or Claude Desktop:
https://mcp.helpcenter.io as the URL.
Already on Growth or Catalyst? The connector URL and these steps are waiting in your help center's Settings, under AI agents.
claude mcp add --transport http helpcenter https://mcp.helpcenter.ioAdds the same connector to Claude Code. You sign in the same way.
Create a key in your help center's Settings → API Keys and send it with each request to the same URL. Keys can be read-only, or read and write.
Access is granted help center by help center, and you can take it back whenever you like. The sign-in runs on OAuth 2.1, on our own consent screen, so your password never reaches the assistant.
Claude signs in through HelpCenter.io over OAuth 2.1. You choose which help centers it can reach and see every permission it asks for before you allow it: read content, write content, analytics, editorial notes.
Change which help centers a connection can reach, or disconnect it, in Account → Connected Apps. Delete an API key in Settings → API Keys and it stops working.
Deleting a category, discarding a staged edit and restoring an old version are labelled destructive, so clients like Claude ask you before they run. A category with content in it stays put until you choose where the content goes.
New articles can be saved as drafts, so publishing stays your call. On Catalyst, an agent can stage an edit to a live article privately and leave it for your team to publish or discard.
Reader email addresses are withheld unless a connection with write access asks for them. IP addresses are never exposed, and private editorial notes never appear in the comment tools.
Search data is aggregated daily. Your agent sees what was searched and how often, never a trail of individual visitors.
No add-on and no charge per tool call. The MCP server comes with the plan, and your 14-day trial includes it.
$119 /mo, billed annually
or $149 month to month
$179 /mo, billed annually
or $199 month to month
Bootstrap and older plans priced below Growth do not include the MCP server. Compare every plan.
Claude on the web, Claude Desktop and Claude Code, plus any MCP client that supports remote servers with OAuth sign-in. For an agent you build yourself, create an API key in your help center's Settings → API Keys and send it to the same URL.
It is included in the Growth and Catalyst plans, and in the 14-day trials of both. Staged changes and version history need Catalyst. Bootstrap and older plans priced below Growth do not include it, and you can compare plans and upgrade from the pricing page.
No. There is no add-on to buy and no charge per tool call: the MCP server is part of the Growth and Catalyst plans. Standard API rate limits apply.
Only if you give it write access. A read-only API key can search, read and report, and change nothing. With write access, ask for drafts and publishing stays your call, and on Catalyst it can stage edits to live articles for your team to publish. Deleting a category, discarding a staged edit and restoring an old version are labelled destructive, so clients like Claude ask you before running them.
Only the help centers you connect, and only what the permissions you approved allow. Reader email addresses are withheld by default, IP addresses are never exposed, and private editorial notes never appear in the comment tools. Search analytics are aggregated daily, so the agent sees questions and demand, not a trail of individual visitors.
Open Account → Connected Apps in HelpCenter.io to change which help centers a connection can reach, or to disconnect it entirely. For an agent of your own, delete its key in Settings → API Keys.
Yes. Create an API key in your help center's Settings → API Keys and send it as Authorization: Bearer <key> to https://mcp.helpcenter.io. Your agent gets the same tools, scoped to that help center. The connection steps walk through it.
Start a 14-day trial, connect Claude, and ask what your customers could not find. No credit card needed.