> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neoagent.io/llms.txt
> Use this file to discover all available pages before exploring further.

# GitBook API

> Search and read your GitBook runbooks, and draft documentation updates in change requests for review

An agent with this tool works with your GitBook documentation: it finds the runbook a ticket needs and answers from it, and it drafts new or updated pages in a change request for a person to review and merge.

> [!NOTE]
>
> Automatically enabled when you configure GitBook permissions in your agent workflow. No manual toggle needed.

## What It Does

* Search every space of your organization, or one space, and read any page as markdown by its path or id
* List a space's pages, read earlier revisions, files and reusable content
* Open a change request, add, update or delete pages in it, ask reviewers, and comment
* Merge a change request, which publishes its pages
* Read docs sites, what visitors asked Ask AI, documentation issues, members, teams and the audit trail

The agent calls GitBook through the [Vendor API](https://docs.neoagent.io/agents/tools/vendor-api.md) tool, with `gitbook` as the vendor id. [Sandbox](https://docs.neoagent.io/agents/tools/built-in/sandbox.md) scripts can call GitBook too, under the same permissions.

## Permission Groups

| Group | Access levels | What it covers |
| - | - | - |
| **Spaces, Pages & Search** | Disabled, Read Only, Read / Write | Spaces, collections, pages, revisions, files, search, OpenAPI specs, translations and imports |
| **Change Requests & Comments** | Disabled, Read Only, Read / Write | Change requests, their page changes, reviewers and comments |
| **Publishing & Export** | Disabled, Read Only, Read / Write | Merging, reviews, publishing and unpublishing a site, Git Sync import and export |
| **Sites, Navigation & Insights** | Disabled, Read Only, Read / Write | Docs sites, structure, customization, redirects, share links, visitor authentication, MCP servers, Ask AI insights and documentation issues |
| **Members, Teams & Access** | Disabled, Read Only, Read / Write | Members, roles, teams, invites, SSO, SAML, and space, collection and site permissions |
| **Organization, Integrations & Audit** | Disabled, Read Only, Read / Write | Organization settings, audit events, AI usage, style guides, integrations and Git Sync installations |

## Access Profiles

> [accordion] Read Only
>
> Every group Read Only. Search and read; no changes.

> [accordion] Helpdesk
>
> Change Requests & Comments and Publishing & Export Read / Write with approval, so a technician approves each draft and each merge. Every other group Read Only.

> [accordion] IT Admin
>
> Every group Read / Write. Spaces and collections change on their own; change requests, publishing, sites, access and organization settings ask a technician first, because a site change reaches every visitor at once and page text can embed another website.

> [accordion] Full Automation
>
> Every group Read / Write with no approval, merges included, except the writes below that always ask.

## Safety

* Page text is written by people and other tools, and the agent is told to treat it as data. What limits a page that tries to steer the agent is the agent's permission groups and approvals: give an agent that reads pages written by others only the groups it needs, with approval on its writes.
* Page text can link or embed another website, and an embedded image loads when someone opens the change request. Helpdesk and IT Admin ask before each change-request write for that reason; Full Automation does not, so choose it only for an agent whose input you trust.
* The agent edits pages through change requests, and nothing in a change request is live until a merge. Applying a template, importing a website or running a translation into a space, and deleting a space, change live content directly; they need Spaces, Pages & Search at Read / Write, which the Helpdesk profile does not give.
* These writes always ask a technician, in every profile: granting or changing access to a space, collection or site; changing a member's role; creating a team or adding team members; inviting people or joining an organization; SSO membership; any change to a SAML provider; creating or changing a share link; changing visitor authentication or its key; adding or changing a site's MCP server or context connection; publishing, unpublishing, installing or configuring an integration; adding or changing a Git Sync installation; transferring a collection; creating an organization or changing its settings, a site's GitBook Agent channels or agent settings; any write that sets an access control (a visibility, a default role, a space's edit mode or merge rules, an adaptive-content condition, SSO, email domains or invite links); any write that sets an address GitBook fetches or a visitor's browser loads (a file added by URL, a website import, an OpenAPI source, a nav link, a logo or favicon); and deleting a site. Removing access, a team member or an invite, and deleting a share link, an MCP server or a context connection, follow the group's approval setting.
* Neo hides credentials GitBook returns: every credential field GitBook's API defines, a share link's address, an invite link's id, a file's download address, an MCP server's headers and an integration's configuration.

## What it does not do

GitBook opens 31 operations of its API only to its own app or to an integration's token, never to a personal access token: email domains, SCIM connections, custom fonts and hostnames, file upload URLs, and an integration's own tokens and tasks. Neo does not send them; a person does these in GitBook.

Neo also creates no temporary preview or PDF link: each opens private content to whoever holds it for an hour. The agent reads pages as markdown instead. It does not browse the GitHub, GitLab, Slack, Jira or Linear accounts behind the connections GitBook stores: it reads no repository, branch or file through Git Sync, discovers no channel scopes and tests no context connection, because those read data outside your documentation. Git Sync import and export (Publishing & Export) and a context connection sync (Sites, Navigation & Insights) do act on the connected provider; they follow those groups' approval settings. It renders no integration's UI and computes no document, because either can run an integration's own code. It resolves no embed, because GitBook would fetch whatever address is named. It does not ask GitBook's AI, because each question spends your organization's GitBook AI usage; it searches and reads the pages itself. And it does not query raw visitor analytics, because an event's address can carry a share link's or a visitor's private token.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.