> ## Documentation Index
> Fetch the complete documentation index at: https://docs.arcdemo.io/llms.txt
> Use this file to discover all available pages before exploring further.

# ArcDemo MCP server

> Connect AI assistants to ArcDemo over the Model Context Protocol with OAuth. Read demos and analytics, draft and apply edits, and publish with admin approval.

Connect AI assistants such as Claude, ChatGPT and Cursor to your ArcDemo workspace over the Model Context Protocol (MCP). Agents can read your demos, analytics and leads, draft and apply edits you can review, and, with an admin's approval, publish demos or sync leads to your CRM.

> [!NOTE]
>
> MCP access is enabled per deployment. The public endpoint is not yet available, so ask [support@arcdemo.io](mailto:support@arcdemo.io) for availability and the server URL for your workspace.

## Connect

* Use the Streamable HTTP server URL provided for your enabled deployment. The server URL is deployment-specific.
* Sign in with OAuth 2.1 (authorization code with PKCE). You choose one workspace and the scopes the client gets. Discovery follows RFC 9728 protected-resource metadata, so MCP clients that support OAuth find the sign-in page on their own.
* Clients without an OAuth flow can use a personal MCP token from *Settings → Agent access*. It is shown once, is bound to you, one workspace and the MCP server, and expires (30 days by default).

## What agents can do

* **Read:** demos, steps, collections, share links, workspace context, analytics, step funnels, visitors, accounts and leads, limited by the scopes you grant.
* **Draft and edit:** audit a demo, draft copy, plan HTML edits, preview a change set, apply it against the current revision, and revert it.
* **With approval:** publishing, unpublishing, creating or revoking share links, and CRM sync need a single-use approval from a workspace admin in ArcDemo. A model saying "confirmed" is never enough.
* There are no delete or admin tools.

The server also offers read-only resources and a few guided prompts. Everything an agent sees is filtered by the same permissions as the tools.

## Safety model

* Access comes from the credential and your membership, never from tool arguments.
* Edits are checked against the demo's current revision, so an agent can't overwrite newer work.
* Captured pages, demo text and CRM data are treated as untrusted content, never as instructions.
* Every call is rate limited and written to the workspace audit log.
* CRM credentials stay in ArcDemo, encrypted. Agents act on a connection you set up in *Settings → Integrations* and never see the provider's tokens.

## REST API for automation

When enabled for your deployment, the agent platform exposes capabilities to backends over HTTPS at `/api/agent/v1/capabilities/{name}`. Use a scoped, expiring service credential created in *Settings → Agent access*; ask support for your deployment’s API origin and availability.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.