MCP Servers
Compare the Reader MCP Server and the Authoring MCP Server, and choose the right one for reading published docs or authoring them.
Two servers, two audiences
Documentation.AI provides two MCP servers. They differ by who connects to them.
The Reader MCP Server is your endpoint, served from your documentation domain. Your readers connect to it to search and reference your published documentation: customers, support teams, internal users, and the AI tools they work in. This server was previously called the Docs MCP Server.
The Authoring MCP Server is Documentation.AI's endpoint. Your documentation team connects to it from an AI coding tool to edit pages, manage branches, and deploy changes.
Compare the servers
| Reader MCP Server | Authoring MCP Server | |
|---|---|---|
| Who connects | Your readers, support teams, and internal users | Your documentation team |
| Endpoint | Your docs domain, for example docs.yourdomain.com/_mcp | api.documentationai.app/mcp |
| Access | Read-only, publicly published content | Read and write |
| Setup | None. The endpoint is live as soon as you publish. | Connect your client, then sign in with OAuth 2.1 |
| Permissions | Public content only. Private, protected, and partial docs are not served yet. | Your Documentation.AI workspace role |
If you are choosing between them, ask who is connecting. If the answer is "someone reading my documentation," use the Reader MCP Server. If the answer is "someone changing my documentation," use the Authoring MCP Server.
Go to the server you need
Reader MCP Server
Give AI tools read-only access to your published documentation.
Authoring MCP Server
Edit docs, manage branches, and deploy from an MCP-compatible AI client.
If you would rather use HTTP
The Documentation.AI REST API covers largely the same documentation-management tasks as the Authoring MCP Server, through a different interface. Use the REST API when you want direct HTTP access from your own scripts, services, or CI/CD, and use the Authoring MCP Server when you want an MCP-compatible AI client to do the work. See the API reference overview.