AgentDock

1.7k
Dock Agent Api
Cursor

Cursor

Use Cursor with the Dock workspace. Add the MCP server to `.cursor/mcp.json` and Cursor's agent can read, write, suggest, and publish documents. CLI works from the embedded terminal.

Cursor's agent speaks MCP. Add the Dock workspace as an MCP server and Cursor can pull prompts from your library, draft documents, propose tracked changes, manage citations, and publish, all from the same chat panel you already use for code.

This is the same workspace Claude Code sees. The agent does not care which client called it; it sees the 38 tools the server exposes.

Cursor reads MCP servers from .cursor/mcp.json in the workspace root, or from the global ~/.cursor/mcp.json. Use whichever scope fits.

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

Reload the window (Cmd+Shift+P -> Developer: Reload Window). The next Cursor agent session sees the AgentDock tools.

Screenshot
Cursor's MCP settings panel with the agentdock server listed and connected.

If you already installed the CLI (next section), the JSON is one command away:

agentdock mcp-setup

The token is filled in for you. Paste it into .cursor/mcp.json.

npm install -g agentdockai
agentdock login

login opens the browser. Approve the device. The CLI stores an ak_usr_* API key under ~/.agentdock/config.json. Same key, same scope as the MCP header.

The CLI is handy from Cursor's embedded terminal when you want to script outside the agent loop:

agentdock docs prompts
agentdock docs list --limit 10
agentdock docs publish <doc-id> --seo-title "Q3 Board Memo" --slug q3-board-memo
Screenshot
Cursor with the agent panel on the right and the embedded terminal running agentdock at the bottom.

Pull a prompt and write inside the workspace

Use the "blog post outline" prompt from AgentDock to draft a piece on
streaming SSE patterns. Save it as a Dock document called
"Streaming SSE patterns draft."

get_user_prompt -> create_document -> write_document. The draft lives in your AgentDock library, separate from any code file in the Cursor workspace.

Round-trip between code and prose

Read the Dock document "Streaming SSE patterns draft." Update the code
snippets to match what I just shipped in `apps/api/src/stream/handler.ts`.
Propose tracked changes; don't overwrite.

read_document -> reads your repo file -> suggest_edits. The doc gets revised proposals; you accept/reject from the AgentDock editor.

Cite from inside Cursor

The doc references the Lewis et al. RAG paper and the GPT-4 technical
report. Add citations to both, in IEEE style.

add_citation (twice) -> update_document_settings to switch the style.

Publish without leaving Cursor

Publish the draft. SEO title: "Streaming SSE Patterns for AI Agents in 2026."
Description: "How to ship reliable streaming responses from agent backends."

publish_document writes the page to /p/slug with that metadata.

Video
60-second walkthrough of all four workflows from inside Cursor.

Cursor's MCP server connection uses the x-api-key header. Same token as the CLI. Every call needs the key today; a keyless tier for the public prompt library is coming soon.

Revoke a token at any time from your AgentDock account settings; the next MCP call will return 401 and you can issue a new one.

See Authentication for token lifecycle.

MCP server shows "disconnected" in Cursor. Cursor caches MCP state per-window. Reload the window after editing .cursor/mcp.json.

401 on the first call. The header is missing or the token expired. Run agentdock login and agentdock mcp-setup, paste the refreshed JSON, reload Cursor.

Cursor agent picks code-editing tools when you want it to use Dock. Be explicit in the prompt: "use the AgentDock MCP server" or "use Dock tools." Cursor prioritizes its own native tools by default.

ClineExamples and Workflows