Documentation.AI

Returns the current content of one file on a working version in the documentation project, including unpublished changes and YAML frontmatter for an MDX page. Use this to read a page before changing it, or to quote what a page says. It returns one file, so use list_pages to discover paths and search to find pages by topic. <examples> get_page(path='docs/getting-started.mdx') get_page(path='documentation.json') get_page(path='docs/guide.mdx', branch='update-auth-docs') get_page(path='api/openapi.yaml', find='x-rate-limit') get_page(path='changelog.mdx', offset=1200, limit=200) </examples> <hints> - path is a file path with its extension, as list_pages reports it. A navigation path such as "docs/guide" will not resolve. - Reads the current content in the editing session, including changes made by other collaborators. - This is the text an edit is matched against, so an edit built from anywhere else may match nothing. - A file over 200 KB is too large to return whole, so its opening lines come back with, for a page, its outline. offset and limit read any part, find locates text, and outline maps a page; edits still apply to the whole file. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#read-content

get_page

Returns documentation.json from a working version in the documentation project, including unpublished changes to navigation, branding, SEO, redirects and integrations — the whole file, one part of it, an outline of its navigation, or the navigation entries matching a search. Use this to understand the site's structure before changing it, to find where a page sits in the navigation, or to read site-wide settings. It does not return page content — get_page does that. <examples> get_site_config() get_site_config(outline=true, depth=2) get_site_config(find='webhooks') get_site_config(pointer='/navigation/tabs/0/groups/3') </examples> <hints> - Reads the current content in the editing session, including changes made by other collaborators. - On a large site, start with outline: every tab and group with its pointer and the number of pages and links beneath it, in a few kilobytes. Then read one part in full with pointer, or locate an entry by title or path with find. - The whole file can only be returned while it is under 200000 bytes. Past that, outline, find and pointer still work. - The pointers returned are the ones update_site_config takes. They are positions, so guard a write with a test patch on the entry’s name: if the navigation changed since it was read, the write is refused rather than applied to the wrong entry. - When the file is not valid JSON, config is null and parseError and raw are returned instead, so the file can be repaired rather than guessed at. - Navigation entries hold navigation paths, without the .mdx extension. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#read-content

get_site_config

Imports up to 10 files into the documentation project from public https URLs, and returns the stored address and the MDX snippet for each. Use this to bring a diagram, screenshot or document into the library so a page can show it. It stores the file and returns its address and snippet; no page is changed. <examples> import_media({ files: [{ url: "https://example.com/architecture.png", altText: "Service architecture" }] }) import_media({ files: [{ url: "https://example.com/a.png", name: "overview.png" }, { url: "https://example.com/b.mp4" }] }) </examples> <hints> - https only, and the address must be reachable from the public internet. - A file whose bytes are already in the library is reused rather than stored twice. - A name already taken gets a suffix, so importing screenshot.png twice gives screenshot-2.png. - One result per file: some may fail while the rest succeed. Check failed. - Give images altText. It is what a screen reader announces. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#edit-content

import_media

Lists unpublished changes in a working version of the documentation project, including files added, modified or deleted. Use this to review what publish would include. It covers changes from all collaborators in the working version, whether made in the editor or through MCP. Publishing the live version updates the live site. <examples> list_changes() list_changes(branch='update-auth-docs') </examples> <hints> - changeStatus is added, modified or deleted. - An empty list means the working version has no unpublished changes; it does not indicate what is on the live site. - diff_page shows the changes inside one file. - Pass nextCursor as cursor to retrieve more results. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#review-and-publish

list_changes

Lists images, video and files stored in the documentation project, with the public address of each and the MDX that places it on a page. Use this to see what a project holds, to get the snippet for a file, or to see what was added after someone uploaded from the dashboard. It reads the library and changes nothing. <examples> list_media() list_media({ kind: "video" }) list_media({ since: "2026-09-18T10:00:00Z" }) </examples> <hints> - Newest first, so the first page is what changed most recently. - since narrows the answer to files stored after a moment you name, such as when you last looked. - The snippet is the MDX that would place a file on a page; listing changes no page. - Pass nextCursor as cursor to retrieve more results; a large library needs several calls. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#read-content

list_media

Lists files on a working version in the documentation project, including unpublished changes, with a file path and size when available. Use this to get exact file paths, to check whether a file already exists before creating it, or to see what a folder contains. It is not a way to find pages by topic — search answers that, and it does not return file contents. <examples> list_pages() list_pages(branch='update-auth-docs') </examples> <hints> - Returns every file, not only pages: documentation.json, OpenAPI specs and images are included. - Paths are file paths, with the extension, which is the form the page tools take. - A file listed here may still be absent from the navigation, in which case the site does not show it. - Pass nextCursor as cursor to retrieve more results. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#read-content

list_pages

Lists organizations and documentation projects accessible to the signed-in user, including organization roles and the IDs accepted by documentation project tools. Use this to find organizationId and documentationId before selecting a documentation project or targeting one on a tool call. The organization role indicates whether the user can read or change its documentation projects. Page content is not included. <examples> list_projects() </examples> <hints> - role is the role in the organization: admin and editor can change content, viewer cannot. - One organization can hold several documentation projects. - IDs are UUIDs; pass organizationId and documentationId together, exactly as returned here. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#projects

list_projects

Lists review requests for the documentation project, with source and target working versions, status, timestamps and a URL for each request. Use this to see which published changes are awaiting review or whether a working version already has an open review request. This tool only lists requests; it does not create, apply or close them. <examples> list_pull_requests() list_pull_requests(state='closed') list_pull_requests(state='all') </examples> <hints> - Defaults to open review requests. - head identifies the source working version; base identifies the target working version. - mergedAt records when the requested changes were applied. It is null if the request was closed without applying them. - Pass nextCursor as cursor to retrieve more results. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#branches

list_pull_requests

Lists working versions of the documentation project, marking the live version, with the latest publication identifier for each and whether direct publication is restricted. Use this to find exact version names before reading, editing or publishing changes, including the live version to target when applying changes to the live site. It does not select a working version. <examples> list_branches() </examples> <hints> - live is true for the live version. Publishing it, or applying changes to it, updates the live site; other working versions do not. - Restricted working versions may refuse direct publication. - Working version names are case-sensitive. Pass them exactly as returned in the `branch` parameter of other tools. - Pass nextCursor as cursor to retrieve more results. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#branches

list_branches

Moves or renames a file on a working version in the documentation project, preserving its content and editing history. Use this to reorganise files or rename a page. Every navigation entry for the file is updated to the new navigation path in the same call. <examples> move_page(path='guide.mdx', destination='docs/guide.mdx') move_page(path='docs/old-name.mdx', destination='docs/new-name.mdx') </examples> <hints> - destination is the full path to move to, not just a new file name. - After deployment, moving a page changes its URL on the live site. Add a redirect in documentation.json to preserve existing links. - Stays unpublished until publish runs. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#edit-content

move_page

Publishes all unpublished changes in a working version of the documentation project. Publishing the live version also updates the live site. list_changes and diff_page show what would be published. Publication includes changes from all collaborators in that working version, whether made in the editor or through MCP. If newer published content conflicts with the edits, nothing is published and the conflicts are returned. <examples> publish(commitMessage='Document the webhook retry behaviour') publish(commitMessage='Fix broken links', branch='update-auth-docs') </examples> <hints> - Collaborators with the documentation project open see a notice naming who published. - Edits made while publishing remain unpublished so they can be included next time. - For conflicts, the file is left untouched and each conflict returns the merged file with the disputed passages. resolve_conflict takes the reconciled file. - Publishing another working version saves its changes without updating the live site. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#review-and-publish

publish

Replaces a stored file with what a public https URL serves, keeping the same address so every page already showing it picks up the new version. Use this to update a screenshot or diagram without editing any page. The replacement must be the same type as the file it replaces. <examples> replace_media({ mediaId: "b1f2...", url: "https://example.com/architecture-v2.png" }) </examples> <hints> - mediaId is the id a file carries when it is stored or listed. - An image can only be replaced by an image, a video by a video. - The previous version is kept, so support can recover it after a mistaken replacement. - Readers may see the old file for a short while after, until caches catch up. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#edit-content

replace_media

Returns a short-lived link where the person can drop files from their own computer into the documentation project, such as a screenshot. A tool call cannot carry a file, so this is how a local file reaches the library. Use this when the person has a file that is not at a public https address. Give them the link and wait for them to say they are done; whatever they drop is then in the library, stored after the requestedAt this returns. <examples> request_upload({ note: "Screenshot of the billing page" }) </examples> <hints> - The link opens in the dashboard and needs the person to sign in; it lasts 15 minutes. - Uploaded files get the same type, size and quota checks as any upload. - The note is shown on the upload page, so say what you are asking for. - A file already at a public https address does not need this; it can be imported directly. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#edit-content

request_upload

Saves reconciled content for a file with a publication conflict in a working version of the documentation project. Use this after publish returns conflicts. ours already includes every change that does not conflict; diff shows each disputed passage, with the newer published content as removed lines and the unpublished edits as added lines. Supply the complete file, starting from ours, with those passages reconciled. The resolution stays unpublished until publish runs again. <examples> resolve_conflict(path='docs/guide.mdx', content='---\ntitle: Guide\n---\n\n# Guide\n\n...') </examples> <hints> - content is the complete reconciled file, including frontmatter for an MDX page. - Resolve each affected file, then call publish again. - A file removed from the newer published content cannot be resolved with this tool. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#review-and-publish

resolve_conflict

Discards every unpublished change in a working version of the documentation project, across all files and all collaborators, restoring its published state. Use this only to abandon all pending work on a working version. To discard edits to a single page, use revert_page. It does not undo changes already published. <examples> revert_all_changes() revert_all_changes(branch='draft-v2') </examples> <hints> - This discards other collaborators' unpublished work as well as your own. Discarded work is not recoverable. - list_changes shows what would be discarded; check it first. - Collaborators with an affected file open see the restored content immediately. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#review-and-publish

revert_all_changes

Discards unpublished changes to one file in a working version of the documentation project, restoring its published content. Use this to discard unwanted edits to a single page before publishing. It does not undo changes already published. <examples> revert_page(path='docs/guide.mdx') </examples> <hints> - Every unpublished change to the file is discarded, including changes from other collaborators. Discarded work is not recoverable. - Collaborators with the file open see the restored content immediately. - diff_page shows what would be discarded. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#review-and-publish

revert_page

Replaces the content of a page on a working version in the documentation project, applying a line-by-line comparison to preserve concurrent edits where possible. Use this when a change is too structural to express as replacements — a reorganised page, a format conversion. For anything smaller, update_page is safer because it touches only what it names. <examples> rewrite_page(path='docs/guide.mdx', content='---\ntitle: Guide\n---\n\n# Guide\n\n...') </examples> <hints> - content is the complete file, frontmatter included. An MDX page needs at least a title in its frontmatter. - The comparison is made against the current content get_page returns and applied in the editing session; concurrent edits may cause changes to be skipped. - If the page moved on too far for the comparison to apply, the call is refused and the current text is returned. - Stays unpublished until publish runs. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#edit-content

rewrite_page

Searches the live site of the documentation project by meaning and keyword, returning each match with its title, file path, URL and snippet. Use this to find pages by topic. Results reflect the last live site deployment: unpublished changes, and working versions other than the live version, are not searched. get_page and fetch return current content, which may differ from a result. <examples> search(query='configure webhooks') search(query='rate limits', limit=5) search(query='authentication', semanticRatio=0) </examples> <hints> - semanticRatio balances meaning against exact wording: 0 matches keywords only, 1 matches meaning only, and the default of 0.75 favours meaning. - Lower semanticRatio when looking for an exact term such as a function or header name. - id and path are the same file path: fetch and get_page both read the page by it. url is where the published page is served, for citing or linking it; the current text of the page comes from fetch or get_page, not from that URL. - Search does not accept `branch` and does not use the session working version. Use list_pages to discover files on another working version. - No results can mean the topic is missing, or that it is not on the live site yet. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#read-content

search

Finds files in the documentation project by name or alternative text, with the public address of each and the MDX that places it on a page. Use this before importing a file, to check the project does not already have it, and to find one file in a library too large to read through. It searches the stored library, not the pages that use those files. <examples> search_media({ query: "architecture" }) search_media({ query: "logo", kind: "image" }) </examples> <hints> - Matches any part of the file name or its alternative text, ignoring case. - Nothing found is an answer, not a failure: import the file, or ask for the upload link. - Newest first, and paged; pass nextCursor as cursor for the rest. - A file whose bytes are already stored is reused on import, so a near-match is worth checking. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#read-content

search_media

Selects the default documentation project and optional working version for later tool calls. The documentation project selection is remembered for seven days; the working version selection lasts for this session. Use this when working on one documentation project across several calls. To target a different documentation project for one call, pass organizationId and documentationId on that call without changing the selection. <examples> select_project(organizationId='11111111-1111-4111-8111-111111111111', documentationId='22222222-2222-4222-8222-222222222222') select_project(organizationId='11111111-1111-4111-8111-111111111111', documentationId='22222222-2222-4222-8222-222222222222', branch='update-auth-docs') </examples> <hints> - Both IDs come from list_projects and are UUIDs. - Omitting working version clears the previous session working version, so later calls default to the live version. - Use list_branches to find working version names. Selecting a working version does not create it. - Selecting an inaccessible documentation project fails and preserves the previous selection. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#projects

select_project

Replaces exact text in a page on a working version in the documentation project. Changes appear in the editing session and stay unpublished until publish runs. Use this for focused changes to an existing page: a sentence, section or code sample. Edits are matched against the current content get_page returns. create_page adds a new page and rewrite_page replaces most of one. <examples> update_page(path='docs/guide.mdx', edits=[{old_string: '## Setup', new_string: '## Getting started'}]) update_page(path='docs/guide.mdx', edits=[{old_string: 'v1.2', new_string: 'v2.0'}], replaceAll=true) </examples> <hints> - old_string must be the exact text from the page, copied with its indentation, and must appear only once unless replaceAll is set. - Read the page first: it may have changed since it was last seen, and an edit is matched against what it says now. - The stored text is not always what was written. Saving normalises a page — underscores in prose become \_, components gain their default attributes such as collapsed="false", long frontmatter values fold onto several lines — so build old_string from what get_page returns, never from what was sent to create_page or rewrite_page. - Several edits apply in order, so a later one can target text an earlier one produced. - If an edit matches nothing the whole call is refused and the page is left alone, with the current text returned to rebuild from. - A skipped change means that passage moved on between reading the page and applying the edit. The other changes still apply, and the skipped one is reported rather than forced. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#edit-content

update_page

Updates documentation.json on a working version in the documentation project by adding, removing, replacing, moving or copying values at JSON pointers. Use this for any change to site configuration — a navigation entry, a colour, a redirect, the site name. get_site_config returns the current structure, and a pointer addresses a place in it. The change is checked before it lands: one that would leave the navigation pointing at a page that does not exist, or otherwise break the site, is refused with the reason. Not for page content, which update_page handles. <examples> update_site_config(patches=[{op: 'add', path: '/navigation/tabs/0/groups/1/pages/-', value: {title: 'Webhooks', path: 'docs/webhooks'}}]) update_site_config(patches=[{op: 'replace', path: '/colors/light/brand', value: '#0c6b56'}]) update_site_config(patches=[{op: 'test', path: '/navigation/tabs/0/groups/1/group', value: 'v4.2'}, {op: 'add', path: '/navigation/tabs/1', value: {tab: 'v4.2'}}, {op: 'move', from: '/navigation/tabs/0/groups/1/pages', path: '/navigation/tabs/1/groups'}, {op: 'remove', path: '/navigation/tabs/0/groups/1'}]) </examples> <hints> - Pointers follow RFC 6901: segments separated by "/", array entries by index, and "-" to append to an array. - Navigation entries carry navigation paths, without the .mdx extension. - Patches apply in order as one atomic change, validated once after the last, so a later patch can address what an earlier one created. - To restructure, use move or copy: they name a source, so a subtree is never resent as a literal. - Guard entries addressed by index with a test on their name, as in the last example: if a collaborator moved one, the whole patch is refused. - A refused change leaves the file untouched and names every problem. A change the site would merely ignore lands, with warnings. - The file is written in its canonical two-space form. Stays unpublished until publish runs. </hints> Documentation: https://documentation.ai/docs/ai/authoring-mcp-server#edit-content

update_site_config