MCP Server Overview

The Autype MCP server implements the current Model Context Protocol authorization model over Streamable HTTP. It exposes Autype’s document generation capabilities as tools that compatible AI assistants and IDEs can call directly — no custom code required.

What is MCP?

The Model Context Protocol (MCP) is an open standard that allows AI agents and applications to access external tools and data sources. Think of it as a universal API that AI assistants understand natively. With the Autype MCP server, your AI agent can:

Generate documents from Extended Markdown and receive a download URL
Create and edit persistent documents in your Autype workspace using Extended Markdown plus compact metadata
Build documents iteratively — create a session, append Markdown, then render
Reuse organization styles — list, create, update, and apply shared style presets by ID
List projects — browse your Autype projects and documents
Upload and reuse media — manage temporary render images and toolkit files
Process PDFs and source files — merge, split, convert, OCR, classify, and extract structured data
Roundtrip Word documents — import DOCX as editable Autype JSON and render the edited result back to DOCX
Use JSON when needed — advanced clients can still render or inspect full document JSON for complete structural control

Connect with OAuth (recommended)

Use this remote MCP URL:

https://mcp.autype.com/mcp

The canonical server display name is Autype. ChatGPT, Codex, Claude, or an IDE may derive a local alias such as autype-prod or autype_prod; that alias belongs to the client UI and is not an additional Autype endpoint. Always use the exact server label exposed by the current client when addressing a tool or resource. In production, Autype discovery, authorization, and MCP resource URLs always use public HTTPS endpoints. Local native clients such as Codex CLI may still register an ephemeral http://127.0.0.1:<port>/callback/... redirect URI. That loopback address belongs to the client on the user’s machine and is not an Autype production endpoint. Hosted clients such as ChatGPT and Claude register their own HTTPS callback URLs instead. OAuth-capable clients discover Autype’s authorization server automatically. The client opens Autype in your browser, where you sign in, choose an organization, review the requested permissions, and approve or deny access. You never copy an API key into the client. Autype supports:

OAuth 2.1 authorization code flow with PKCE (S256)
Protected Resource Metadata and authorization server discovery
Client ID Metadata Documents (CIMD) and Dynamic Client Registration (DCR)
Resource-bound, short-lived access tokens
Server-side token exchange: MCP access tokens are never forwarded to the Developer API
Rotating refresh tokens when the client requests offline_access
Per-tool scopes and security metadata

Your Autype plan must include Developer API access. OAuth grants are bound to the selected organization and can never expand beyond the permissions approved during consent.

Manage or disconnect apps

Open Autype Settings → Connected Apps to review every active OAuth connection. Each entry shows the client, workspace, granted permissions, and whether the client has background access. Connections are personal: removing your connection does not disconnect another member who authorized the same client. Disconnecting an app immediately revokes all of its access and refresh tokens for the selected workspace, as well as any authorization code that has not yet been exchanged. The OAuth client registration itself remains available so the app can be connected again through the normal consent flow.

Legacy API-key clients

Clients without OAuth support can continue to send an Autype API key in the X-API-Key header. Existing configurations remain compatible, but OAuth is preferred for interactive ChatGPT and Claude integrations because credentials are short-lived, scoped, revocable, and never stored in client configuration files.

Supported clients

ChatGPT

Connect Autype as a custom app in ChatGPT using OAuth.

Claude

Connect Autype as a remote connector in Claude using OAuth.

Claude Desktop

Connect Autype to Claude Desktop app for macOS and Windows.

Claude Code (CLI)

Use Autype with Claude’s command-line interface.

Cursor

Integrate Autype into the Cursor AI-powered code editor.

Windsurf

Connect Autype to Windsurf IDE with Cascade.

VS Code (Copilot)

Use Autype with GitHub Copilot in Visual Studio Code.

MCP Inspector

Test and debug the Autype MCP server with the official inspector tool.

OAuth permissions

Scope	Allows
`read:documents`	Read and list documents, projects, records, styles, blocks, and render status.
`write:documents`	Create or update documents, records, builder sessions, and related content.
`render:documents`	Start document and image renders.
`manage:resources`	Create, update, or delete reusable styles and blocks.
`manage:files`	Access file-related Developer API operations.
`offline_access`	Let the client obtain a rotating refresh token; it is not included in access-token scopes.

Autype validates the required scope again for every MCP tool and Developer API request. If a tool needs a permission that was not granted, the request is rejected instead of silently broadening access.

Available tools

Recommended authoring model

For MCP agents, the preferred format is Autype Extended Markdown (AEM) plus compact JSON metadata. AEM is Autype’s semantic document language, not merely plain Markdown:

Put readable content in Markdown.
Pass variables, abbreviations, citations, document settings, and style_preset_id as separate fields.
Use workspace or built-in system style presets by ID instead of sending large inline defaults objects.

The full Autype JSON schema remains supported for advanced clients through explicit JSON render/retrieval tools. The builder tools are Markdown-only so agents do not have to choose between two authoring models.

Persistent document tools

Tool	Description
`documents_list`	List persistent documents in the organization, optionally filtered by project.
`documents_create_markdown`	Create a saved Autype document from Extended Markdown. This is the recommended persistent create flow for MCP agents.
`documents_create_json`	Advanced/migration: create a saved document from the complete Autype JSON schema.
`documents_get_markdown`	Read a saved document as Extended Markdown plus compact metadata (`document`, `variables`, `abbreviations`, `citations`, `stylePresetId`, etc.).
`documents_update_markdown`	Update a saved document from Extended Markdown. Omitted metadata is preserved by the API.
`documents_get_markdown_outline`	Return heading/directive chunks, line ranges, hashes, and revision for a saved Markdown document.
`documents_get_markdown_chunk`	Retrieve one Markdown chunk by `chunkId` or a direct `line_start`/`line_end` range.
`documents_patch_markdown`	Apply line-based Markdown patch operations with optional revision/hash conflict checks.
`documents_get_json`	Advanced/debug: read the full internal Autype JSON snapshot.
`documents_render`	Render the latest saved snapshot to PDF, DOCX, ODT, PNG, or JPEG.

Render tools

Tool	Description
`render_document`	Render AEM to PDF/DOCX/ODT. Takes `content` and `document` as separate top-level parameters, plus optional `style_preset_id`, `defaults`, `variables`, `abbreviations`, and `citations`. This is the recommended default for AI agents.
`render_json`	Advanced/debug: render a complete document JSON to PDF/DOCX/ODT. Takes a `config` object with `document` and `sections`.
`render_document_images`	Render Extended Markdown to PNG/JPEG page images for visual QA. A single page is returned directly; multiple pages are a ZIP.
`render_json_images`	Advanced/debug: render complete document JSON to PNG/JPEG page images.
`check_export_readiness`	Strictly validate semantic AEM source readiness without starting a render job or consuming credits. Its score is not visual approval; inspect rendered pages and every requested final format separately.
`render_get_status`	Poll for render job status by job ID. Returns status, download URL (when completed), format, credit cost, and timestamps.
`render_list_jobs`	List render jobs with pagination (`page`, `limit`) and optional `status` filter (“PENDING”, “PROCESSING”, “COMPLETED”, “FAILED”).

The MCP render tools enable strict validation by default. Redundant page breaks, broken internal references, and other readiness errors are returned before a job consumes credits. Set strict=false only when intentionally rendering a legacy document that depends on permissive behaviour.

Style tools

Tool	Description
`styles_list`	List compact reusable style summaries. Use `source=workspace`, `source=system`, or `source=all`; every result includes its source and whether it is editable.
`styles_get`	Read one workspace or system style, including the reusable defaults definition. Built-in IDs use `system:<id>` and are immutable.
`styles_create`	Create a new organization style preset.
`styles_update`	Update a style preset name, description, default flag, or definition.
`styles_delete`	Delete an organization style preset.

Reusable block tools

Tool	Description
`reusable_blocks_list`	Search organization blocks by text or category.
`reusable_blocks_get`	Read a block’s current Extended Markdown and version history.
`reusable_blocks_create`	Create a reusable block from Extended Markdown. Use only when the user explicitly wants to manage the library.
`reusable_blocks_update`	Update metadata or publish a new Markdown version.
`reusable_blocks_delete`	Delete a library block; existing references retain fallback content.
`reusable_blocks_markdown_directive`	Build the correct reference or snapshot directive for a known block ID.

Record tools

Tool	Description
`records_list`	List saved customer/case records for a document.
`records_prepare`	Compare candidate values with document variables and report missing, unused, and type-mismatched values.
`records_create`	Save a version-aware variable dataset without rendering it.
`fill_template_variables`	Validate values and create a record only when required values and types are compatible.
`records_get`	Read one record and its recent export history.
`records_update`	Update record values, label, metadata, or status.
`records_delete`	Archive a saved record while preserving its audit history.
`records_render`	Render a saved record with optional one-off value overrides.

Document builder tools

Tool	Description
`builder_create_session`	Create a new session. Accepts `document_type` (“pdf”/“docx”/“odt”), `size` (“A4”/“A3”/“A5”/“Letter”/“Legal”), `orientation`, `title`, and margin settings. Returns a `sessionId`.
`builder_get_session`	Get the session state: Markdown content, logical Markdown sections, metadata, variables, abbreviations, citations, and style preset ID.
`builder_list_sessions`	List all active builder sessions.
`builder_update_document`	Update document-level settings (type, size, orientation, title, author, filename, margins). Only provided fields are changed.
`builder_set_markdown`	Replace the session content with Extended Markdown. Prefer this over JSON sections for agent-authored documents.
`builder_append_markdown`	Append Extended Markdown to the session. Useful for building longer documents step by step.
`builder_add_markdown_section`	Add a logical Markdown section. Sections are flexible: chapters, clauses, appendices, page sections, or any split chosen by the agent.
`builder_update_markdown_section`	Replace or rename one logical Markdown section.
`builder_get_markdown_section`	Retrieve one logical Markdown section by ID.
`builder_remove_markdown_section`	Remove one logical Markdown section.
`builder_reorder_markdown_sections`	Reorder logical Markdown sections by ID.
`builder_get_outline`	Return heading/directive chunks, line ranges, and hashes for the builder Markdown content.
`builder_get_chunk`	Retrieve one builder Markdown chunk by `chunkId` or line range.
`builder_patch_chunk`	Apply line-based Markdown patch operations to builder content.
`builder_set_style_preset`	Set or clear a reusable workspace or built-in system style preset for the session.
`builder_update_defaults`	Set or merge inline defaults for one-off overrides. Prefer `builder_set_style_preset` for reusable styles.
`builder_set_variables`	Set or merge template variables for `\{\{varName\}\}` substitution. Values can be strings, number objects, image objects, list objects, or table objects.
`builder_render`	Render the session to the configured output format. Waits for completion and returns a download URL.
`builder_render_images`	Render the current session to PNG/JPEG pages for visual QA before final export.
`builder_delete_session`	Delete a session and free its memory.

Template tools

Tool	Description
`templates_list_markdown`	Search one bounded template collection at a time (maximum 20 items). `scope=workspace` is the productive default and returns only organization-owned templates. `scope=catalog` browses built-in quick-start templates without mixing them into productive results.
`templates_get_markdown`	Read one accessible template as AEM plus compact document metadata. Check `productive` before using it for an export.
`templates_fill_and_render`	Fill typed variables in an organization-owned workspace template and submit a PDF, DOCX, or ODT render. Built-in catalog templates are intentionally rejected. Poll the returned job with `render_get_status`.

System document templates are designed primarily as frontend quick starts. MCP clients may browse them for orientation through scope=catalog, but should not treat them as organization-approved business templates. Productive template automation always starts from scope=workspace. Styles intentionally differ: both organization styles and built-in system styles are first-class MCP resources, with explicit source markers and filters.

Other tools

Tool	Description
`projects_list`	List projects with pagination (`page`, `limit`). Each project contains document IDs for persistent rendering.
`projects_create`	Create an organization-wide public project for persistent documents.
`auth_get_context`	Show the authenticated organization, represented user, granted scopes, and available credits.

Bulk render tools

Tool	Description
`bulk_get_document_variables`	Inspect typed variables required by a persistent document.
`bulk_render`	Render 1–100 variable sets as a ZIP of PDF, DOCX, or ODT files.
`bulk_render_file`	Upload CSV, Excel, or JSON variable data and create a bulk render job.
`bulk_get_status`	Poll progress and receive the signed ZIP download URL.

Media and file toolkit

Tool	Description
`images_upload`, `images_list`, `images_delete`	Manage 24-hour render images and use returned `refPath` values in AEM.
`files_upload`, `files_list`, `files_get`, `files_delete`	Manage 60-minute source/output files. Binary content is not echoed into LLM context.
`docx_import`	Convert an uploaded DOCX to validated editable Autype Document JSON, including protected image fallbacks and structured quality diagnostics.
`pdf_process`	Merge, split, rotate, keep/remove pages, watermark, inspect metadata/forms, protect/unlock, compress, convert pages to images, fill, or flatten PDFs.
`files_convert`	Convert DOCX, ODT, HTML, and images, or replace typed DOCX/ODT placeholders.
`lens_process`	OCR to Markdown/AEM/JSON, generate filenames, classify documents, or extract typed data.
`tools_get_job`, `tools_list_jobs`	Poll asynchronous toolkit jobs and retrieve output file metadata/download URLs.

Completed toolkit jobs return both outputFileId and an authenticated Autype API downloadUrl. The URL includes a short-lived token scoped to that output file, so MCP clients can fetch it directly without forwarding their bearer token. For images, upload actual PNG/JPEG/WebP bytes with images_upload and insert its returned refPath. The autype-ai-image placeholder is not a valid final MCP render source and strict readiness rejects it. images_upload accepts PNG, JPEG, and WebP as plain Base64, whitespace-wrapped Base64, or a Base64 data URL, up to 25 MB decoded. Images referenced by a persistent document remain available with that document. Toolkit uploads accept PDF, DOCX, ODT, Markdown, HTML, PNG, JPEG, GIF, BMP, and SVG; they remain temporary and expire after 60 minutes. The dispatcher tools publish their complete operation/option contract in the tool description and reject missing, unknown, mistyped, or out-of-range options before a job is queued.

Schema resources

The MCP server exposes the Autype document schema as readable resources. Your AI agent can read these to understand the exact structure before constructing a document:

Resource URI	Description
`autype://guides/authoring`	Token-efficient AEM workflow and complete syntax-family map. Recommended first resource.
`autype://schemas/markdown-syntax`	Complete AEM syntax, attribute, and example reference. Load for advanced syntax.
`autype://schemas/document`	Advanced/debug: full document JSON schema (sections, elements, defaults, variables, etc.)
`autype://schemas/defaults`	Defaults schema (fonts, colors, spacing, header/footer)
`autype://schemas/document-settings`	Document settings (page size, orientation, margins, output format)
`autype://schemas/variables`	Template variables for text substitution
`autype://schemas/abbreviations`	Abbreviation entries for automatic expansion
`autype://schemas/citations`	Citation entries and bibliography configuration
`autype://schemas/style-bundle`	Focused reusable typography/page-design style contract
`autype://schemas/page-style-entry`	Default, first, odd, even, section, and blank page variants
`autype://schemas/page-region`	Multi-row/multi-column header and footer regions
`autype://schemas/region-block`	Text, image, field, rule, QR, spacer, and shape blocks
`autype://schemas/style-operations`	Compact atomic updates for existing reusable styles

Example prompts

Once connected, you can ask your AI agent naturally:

“Generate a two-page PDF invoice for client Acme GmbH, total €2,400, due 2026-03-15. Include a line items table.”

“Convert this markdown report to a PDF with A4 size and 3cm margins.”

“Create a DOCX contract template with {{client_name}}, {{start_date}}, and {{monthly_fee}} variables.”

“Find the approved legal disclaimer block, insert it as a live reference, and create a PDF form with customer and signature fields.”

“Prepare a record for customer Acme, report any missing values, then render it as PDF.”

“List my recent render jobs and download the latest completed one.”

Overview

Integrations

LLM Resources

MCP Server Overview

What is MCP?

Connect with OAuth (recommended)

Manage or disconnect apps

Legacy API-key clients

Supported clients

ChatGPT

Claude

Claude Desktop

Claude Code (CLI)

Cursor

Windsurf

VS Code (Copilot)

MCP Inspector

OAuth permissions

Available tools

Recommended authoring model

Persistent document tools

Render tools

Style tools

Reusable block tools

Record tools

Document builder tools

Template tools

Other tools

Bulk render tools

Media and file toolkit

Schema resources

Example prompts

​What is MCP?

​Connect with OAuth (recommended)

​Manage or disconnect apps

​Legacy API-key clients

​Supported clients

ChatGPT

Claude

Claude Desktop

Claude Code (CLI)

Cursor

Windsurf

VS Code (Copilot)

MCP Inspector

​OAuth permissions

​Available tools

​Recommended authoring model

​Persistent document tools

​Render tools

​Style tools

​Reusable block tools

​Record tools

​Document builder tools

​Template tools

​Other tools

​Bulk render tools

​Media and file toolkit

​Schema resources

​Example prompts

What is MCP?

Connect with OAuth (recommended)

Manage or disconnect apps

Legacy API-key clients

Supported clients

OAuth permissions

Available tools

Recommended authoring model

Persistent document tools

Render tools

Style tools

Reusable block tools

Record tools

Document builder tools

Template tools

Other tools

Bulk render tools

Media and file toolkit

Schema resources

Example prompts