Skip to main content
Studiograph exposes an MCP (Model Context Protocol) server at your workspace’s /mcp endpoint. Any MCP-compatible tool — Claude Code, Codex, Claude Desktop, claude.ai, Cursor — can search, read, and edit your workspace through it, with the same folder access you have in the app. There are two ways to authenticate, and which one you want depends on whether a browser is available: Either way, the connection acts as you — the connected tool sees exactly the folders you can see, nothing more.

Find your endpoint URL

In the web app, open Settings → Agents. The MCP Access section shows your workspace’s endpoint URL (it ends in /mcp) with a copy button. This page is available to everyone, not just owners.

Mint a token (for headless clients)

1

Open Settings → Agents → Advanced

The Advanced disclosure below MCP Access holds your personal access tokens.
2

Create a token

Click New token, give it a name (for example “claude-code”), and create it. The token starts with sg_ and is shown once, together with a ready-to-paste client config — copy both now.

Claude Code

For an interactive session, add the server over HTTP and sign in through the browser:
Then run /mcp inside Claude Code and pick studiograph to complete the browser sign-in. For a headless session — CI, claude -p, or any machine where the browser flow isn’t an option — pass a personal access token instead:
To share the server with everyone working in a project, commit a .mcp.json at the project root (keep the token out of it — use an environment variable):
A one-off headless run can also load the config directly:

Codex

Add the server to ~/.codex/config.toml. Recent Codex versions connect to HTTP servers directly:
Export STUDIOGRAPH_MCP_TOKEN=sg_your_token_here in the shell that launches Codex. If your Codex version doesn’t support URL-based servers, spawn the Studiograph CLI as a stdio server instead (requires the CLI and a local workspace — see Local workspace below):

Claude Desktop and claude.ai

Both use the browser sign-in — no token needed:
  1. Copy the endpoint URL from Settings → Agents → MCP Access.
  2. In Claude, add a custom connector (or “MCP server”) and paste the URL.
  3. Approve access in the browser window that opens.

Cursor and other MCP clients

Any client that accepts a streamable-HTTP MCP server works with the same config shape:
This is exactly the snippet Studiograph hands you when you mint a token — with your workspace URL already filled in.

Local workspace (stdio)

If you run Studiograph locally, clients can spawn the CLI as a stdio MCP server instead of calling a URL:
studiograph mcp runs against the workspace in its working directory (or a path passed as an argument). When a local server is running for that workspace, it transparently proxies to it — so edits made over MCP merge live into any open editor. When no server is running, it falls back to operating on the workspace directly. Either way STUDIOGRAPH_MCP_TOKEN is required, and the same per-user folder access applies.

What connected tools can do

Around 53 tools are available, covering most of what you can do in the app:
Permanently deleting a folder is a human-only action in the web app; no MCP tool can do it.

Revoking access

Your minted tokens are listed under Settings → Agents → Advanced. Open a token and revoke it there; any client using it stops working immediately. The raw token value is never shown again after minting — if you lose it, revoke and mint a fresh one.