MCP Server Setup

Add the Dock workspace to Claude Code, Cursor, or any MCP client. Streamable HTTP transport, 38 tools across eight categories, and the SKILL.md the agent reads first.

If your agent runs in an MCP client, it already speaks the protocol. Add the Dock MCP server once and the whole workspace shows up as tools the agent can call directly: read and write documents, propose tracked changes, manage citations, browse versions, and publish.

MCP is the Model Context Protocol, the open standard Anthropic introduced for connecting AI agents to tools and data. The difference from a plain API is who does the integration work: an API needs custom code per consumer, while an MCP server describes its tools in a standard way, so any MCP client, Claude Code, Cursor, Cline, and the rest, can use them with a config entry instead of an integration project. Dock exposes its whole document workspace as one such server, which is why the setup below is a JSON block and not an SDK.

The server uses streamable HTTP transport: it is a URL, not a local process. There is nothing to install and no stdio bridge to run. The MCP client connects directly to:

https://app.agentdock.ai/api/mcp

Add the server to your MCP client settings. The key goes in the x-api-key header.

{
  "mcpServers": {
    "agentdock": {
      "type": "http",
      "url": "https://app.agentdock.ai/api/mcp",
      "headers": {
        "x-api-key": "ak_usr_YOUR_TOKEN"
      }
    }
  }
}

This is the config for Claude Code (~/.claude.json) and the same shape works for Cursor and other MCP clients. Or let the CLI print it for you:

agentdock mcp-setup

That command outputs the JSON above with your stored token already filled in.

Every call to the server requires a valid x-api-key today. Set the header and all 38 tools, documents, citations, versions, folders, prompts, and publishing, are available.

Coming soon: a keyless tier for the prompt library. The public prompt library will be reachable without a key, so an agent can browse and pull prompts with no setup. Until it ships, keep the header on every request.

Tools are grouped into eight categories. Tool names are the exact call names an agent uses.

Documents (8)

Tool	Does
`list_documents`	List documents, newest first
`read_document`	Read content as markdown or JSON
`create_document`	Create a new blank document
`write_document`	Update content by sending markdown
`delete_document`	Permanently delete a document
`duplicate_document`	Copy with a (Copy) suffix
`move_document`	Move to a folder or back to the root
`import_web_page`	Import a URL as an editable document

Publishing (2)

Tool	Does
`publish_document`	Publish to `/p/slug` with SEO metadata
`unpublish_document`	Revert a published page to a draft

Suggestions (4)

Tool	Does
`suggest_edits`	Propose tracked changes by find and replace
`list_suggestions`	List suggestions waiting for review
`accept_suggestion`	Apply a proposed change
`reject_suggestion`	Discard a proposal without applying it

Folders (4)

Tool	Does
`list_folders`	List all folders
`create_folder`	Create a new folder
`update_folder`	Rename or move a folder
`delete_folder`	Delete a folder (its documents move to the root)

Versions (2)

Tool	Does
`list_versions`	List version history metadata
`read_version`	Read a specific version

Prompts (2)

Tool	Does
`list_user_prompts`	List personal and system prompts
`get_user_prompt`	Read a prompt by ID

Settings (1)

Tool	Does
`update_document_settings`	Set the citation style per document

Citations (15)

In a document:

Tool	Does
`add_citation`	Add a citation by DOI, URL, PMID, or arXiv ID
`search_citations`	Search academic sources
`list_citations`	List the citations in a document
`remove_citation`	Remove a citation from a document

Library:

Tool	Does
`citation_library_list`	List saved citations
`citation_library_save`	Save a citation for reuse
`citation_library_get`	Read a saved citation by ID
`citation_library_search`	Full-text search the library
`citation_library_delete`	Delete a saved citation
`citation_format`	Format a citation in a style (preview, no save)

Collections:

Tool	Does
`citation_collection_list`	List collections
`citation_collection_create`	Create a collection
`citation_collection_add`	Add a citation to a collection
`citation_collection_remove`	Remove a citation from a collection
`citation_collection_delete`	Delete a collection

That is 8 + 2 + 4 + 4 + 2 + 2 + 1 + 15 = 38 tools.

The server ships a SKILL.md the agent reads before it writes: the constraints, the house style, the document templates, and what counts as ready to publish. So the agent knows when and how to use the tools, not just that they exist. See the Overview for what the knowledge layer covers.

Authentication: token types and scopes
Writing Modes: direct write vs suggest mode
Examples and Workflows: MCP recipes for real agents

Getting Started OpenAI Codex

MCP Server Setup

What is MCP?

Transport

Config

Authentication

The 38 tools

SKILL.md

Next steps