Model Context Protocol / v0.1

A document workspace an external AI can actually use.

Docs Studio MCP exposes the real loop behind the app: parse a brief, draft onto a document, inspect blocks, propose reviewable edits, and export the result.

Official TypeScript SDK · persistent DynamoDB workspace · Streamable HTTP

01External AIasks with tools
JSON-RPC
02Docs Studio MCPguards the document loop
state + files
03Reviewable canvasnothing changes by surprise
Tools take action. Resources describe state. Prompts set the working rhythm.
25document tools
03read-only resources
03workflow prompts
02transport modes

THE EXPOSED SURFACE

Small primitives. A complete document loop.

The contract is intentionally close to what the web app already does. Drafting applies the first version; later changes are proposals with an explicit accept or reject step.

01

get_capabilities

Return the server contract, transports, and honest session boundary.

02

create_document

Open a blank Letter or Legal workspace, optionally seeded with HTML.

03

list_documents

List persistent documents belonging to the authenticated workspace.

04

read_document

Return HTML, plain text, blocks, word count, brief context, and pending edits.

05

parse_brief

Extract tasks, objectives, constraints, learning outcome, and rubric weights.

06

draft_document

Run the same streaming draft endpoint used by the Docs Studio canvas.

07

chat_document

Use the document copilot; AI changes return as reviewable proposals.

08

propose_edit

Create a named proposal without mutating the current document.

09

accept_edit

Apply a proposal only after the source document still matches.

10

reject_edit

Close a proposal and leave the document untouched.

11

insert_math

Add an editable MathJax equation to the document.

12

insert_table

Add an editable HTML table with a header row.

13

get_history

Return recent document events for review and audit.

14

export_document

Return HTML, DOCX, or PDF as content/base64.

READ-ONLY CONTEXT

Give the model a map before it edits.

docs://workspace
Authenticated workspace

List persistent documents for the signed-in owner.

docs://document/{documentId}
Document state

HTML, text, blocks, brief, and pending review proposals.

docs://history/{documentId}
Audit trail

Created, drafted, proposed, accepted, rejected, inserted, exported.

THE IMPORTANT PART

AI can move quickly without hiding the decision.

Every mutating path is visible in the API. The review boundary is part of the protocol, not a promise in the UI copy.

1draft_documentfirst draft applies
2propose_editchange waits here
3accept_editor reject_edit

CONNECT IN A MINUTE

Connect an agent to your own documents.

Sign into Docs Studio with Google and create a personal MCP credential. Any client that supports Streamable HTTP can use it; the credential sees only that user's persistent workspace.

STDIO / LOCAL DEVELOPMENTlocal
{
  "mcpServers": {
    "docs-studio": {
      "command": "npx",
      "args": ["tsx", "src/mcp/server.ts"],
      "cwd": "/path/to/docs-studio",
      "env": {
        "DOCS_STUDIO_URL": "http://localhost:9003"
      }
    }
  }
}

Or run npm run mcp:stdio from the repository.

STREAMABLE HTTP / ANY MCP CLIENTremote
{
  "mcpServers": {
    "docs-studio": {
      "url": "https://docss.studio/api/mcp",
      "headers": {
        "Authorization": "Bearer dsm_YOUR_PERSONAL_TOKEN"
      }
    }
  }
}

The token is created once from the signed-in workspace, stored hashed, and can be revoked. It starts with all currently available MCP permissions.

WHAT IS TRUE TODAY

The useful boundary is documented too.

This is a working integration, not a fictional “agent platform”. Here is where the implementation is complete and where the next product work belongs.

Available nowThe same document loop

Create → draft → read → propose → accept/reject → export.

Available nowRemote authenticated workspace

Streamable HTTP at https://docss.studio/api/mcp with a personal Bearer credential.

Available nowReview boundary

Consequential edits can remain proposals until an agent or user explicitly accepts them.

Still explicitLive browser sync

An agent works on persistent documents, not silently on the cursor or selection in an open browser tab.

THE SOURCE IS OPEN

Start with a brief, a blank page, or the protocol.

Docs Studio is a real editor with an external surface now. Read the implementation, open the app, or connect your own AI client.