MCP servers for your docs, DeveloperHub
MCP servers · for agents and readers

Let AI agents
write your docs,
and read them.

Two MCP servers, one for each side of your docs: the Editor MCP lets your team’s AI agents write and publish pages, and the Reader MCP lets your readers’ AI tools search and read what you have published.

In the MCP RegistryDrafts until you publishEditor MCP on every plan

Want the agent that works inside DeveloperHub instead? That is the AI Agent →

claude code / acme-docs Editor MCP
>Add a note on token expiry to the Authentication page
developerhub · find_page_by_slug[1842] Authentication (guides/authentication)
developerhub · get_page# Authentication · 31 more lines
developerhub · edit_pageSaved draft of [1842] Authentication. Use publish_page to make it live.
✓ Added a token expiry callout. It is saved as a draft.
readers still see the published page
Product tour

Every way to edit,
in 42 seconds.

First, Claude Code adds a callout through the Editor MCP, saved as a draft. Then the same docs in Git, and in the visual editor.

01 · For your team

The Editor MCP: your agent, signed in as you.

Work on your docs from Claude Code, Claude, Cursor, Codex or any other AI client that supports MCP. Everyone connects as themselves, so their agent can reach exactly what they can, and nothing more.

claude mcp add --transport http developerhub https://ai.developerhub.io/mcp
That is all it takes in Claude Code. Using another client? The setup guide has it.
✓ Admins choose which projects agents can work in
✓ Everyone signs in with their own account and role
✓ Listed in the official MCP Registry
✓ On every plan, Free included
Find
It knows its way around.

Your agent finds the right page and reads it before it writes a word.

Write
New pages and rewrites.

Ask for a new page, a rewrite or a quick fix. It lands as a draft for your team to review.

Publish
Live when you say so.

When you are happy, ask it to publish a page, a section or a whole version.

API references
Keep the spec current.

Hand it a new OpenAPI spec and your API reference is updated as a draft, ready for review.

Upkeep
Broken links, fixed.

Ask it to find the broken links across your docs, and to fix them while it is there.

What to write next
See what readers missed.

It reads what readers searched for and could not find, and what they said about your pages.

Search analytics and written feedback come with Pro and Enterprise.Everything it can do →

02 · You stay in control

Nothing goes live until someone publishes.

Every page an agent writes or rewrites waits as a draft. Your team reads it in the editor, tweaks what needs tweaking, and publishes when it is ready. Your readers never see a half-finished change.

✓ Drafts sit in the editor like any other change
✓ An agent has the same role as the person who connected it
✓ Need an agent that can never publish? Give it a key that cannot
guides / authentication Draft
Authentication

Every request to the Acme API carries an API key. Keys belong to one project, and you can rotate them at any time.

Token expiry
Keys never expire on their own. A rotated key stops working 24 hours after you rotate it, and a revoked key stops at once.

Send the key as a bearer token in the Authorization header of every request.

added by your agent · not live yetPublish
03 · Automations

Let your pipelines write docs too.

Create an MCP key for a release pipeline, a scheduled job or a script, and it can work on your docs without anyone signing in.

✓ Choose exactly what each key can do
✓ Set when it expires
✓ Revoke it whenever you like
Release notes bot
MCP key · used by your release pipeline
✓Can find, read and write drafts
✗Cannot publish or delete
↻Expires in 90 days
On every release
Draft the release notes.

Your release pipeline asks the agent to write up what shipped. The notes arrive as a draft, ready for someone to read and publish.

Every night
Catch broken links.

A nightly job checks your docs for broken links, so you hear about them before a reader does.

Every week
List what readers could not find.

A weekly job collects the searches that came back empty and turns them into your next pages to write.

Search analytics come with Pro and Enterprise.

04 · Written right

It already knows how your docs are written.

Your agent learns DeveloperHub’s format from the server itself, and everything it writes is checked, so nothing gets lost on the way in. Callouts, code blocks, tabs and tables come out the way your writers would make them.

✓ No format guide to paste into your prompt
✓ Anything that would not show up is flagged to the agent
✓ Working in Git instead? Our free agent skills teach it the same
guides / quickstart Draft
Make your first request

Create an order with one call. Use your test key while you build.

cURLPythonNode
curl -X POST https://api.acme.com/v1/orders \
  -H "Authorization: Bearer $ACME_KEY"
Rate limits
Test keys allow 100 requests a minute. Live keys allow 1,000.
✓ written by your agent · formatted and checked
05 · For your readers

The Reader MCP: your docs, in their AI tools.

Your readers build against your product with an AI assistant open. Switch the Reader MCP on and their assistant can search your docs and read any page or API endpoint, straight from the source.

✓ Readers connect from your docs, in Claude Code, Codex, VS Code or Cursor
✓ Answers come from what you published, not from old copies on the web
✓ Listed in the official MCP Registry
claude mcp add --transport http developerhub-docs https://docs.developerhub.io/mcp
Try it on our own docs, in Claude Code or any MCP client.
reader’s agent / docs…/mcp Reader MCP
>How do I hide a page from some readers?
developerhub-docs · searchWhy to hide a page? · Setting Page Audience · Publishing States
developerhub-docs · read_pagesupport-center/conditional-content
✓ Answered from the published docs, with the pages linked.
Only what you meant to share.

It sees what any visitor sees. Pages you have not published, or have kept for certain readers, stay out of sight.

Included with the AI Assistant.

The Reader MCP comes with the AI Assistant on Pro and Enterprise, for public docs. Private docs? Talk to us.

06 · Also built in

Ready for the AI that just reads the page.

Not every AI tool connects to a server. Most simply fetch a page, and your docs are ready for them too.

llms.txt
A map of your docs for AI.

Every page and every API endpoint, listed for AI tools and kept up to date for you.

Markdown
Every page, ready for AI.

Each page also comes as clean Markdown, so assistants such as ChatGPT, Claude and Perplexity read your content, not your layout.

WebMCP
Tools in the reader’s browser.

Your docs offer tools to AI agents working in the reader’s browser. It is brand new, and browsers are only starting to support it.

Questions

MCP servers, answered.

Does an AI agent publish on its own?

No. What an agent writes waits as a draft until someone publishes it. For automations, you can create keys that cannot publish at all.

Which AI clients work?

Claude Code, Claude, Cursor, Codex and any other AI client that supports MCP. Connecting Claude Code takes one command.

How do I control what an agent can do?

An agent can only do what the person who connected it can do, and only in projects where an admin has switched it on. MCP keys go further: you choose what each key can do and when it expires.

What is the difference between the two servers?

The Editor MCP is for your team: their agents write, edit and publish your docs. The Reader MCP is for your readers: their AI tools search and read what you have published.

Is it in the official MCP Registry?

Yes, both servers are. The Editor MCP is listed as io.developerhub/editor and the Reader MCP as io.developerhub/docs.

Which plans include it?

The Editor MCP and MCP keys are on every plan, Free included. The Reader MCP comes with the AI Assistant on Pro and Enterprise, for public docs.

See the full plan comparison →

Bring your own agent.
Keep the publish button.

Fourteen days free, no card. Switch the Editor MCP on, connect your AI client, and ask it for your first draft.