Skip to main content
Connect your AI tools to Architect using the Model Context Protocol (MCP), an open standard that lets AI assistants interact directly with your Architect workspace.

What is Architect MCP?

Architect MCP is Architect’s official MCP server. It’s a hosted, remote MCP with OAuth, so you don’t need to run anything locally. It’s available at: https://mcp.tryarchitect.com/mcp It integrates with AI assistants like Claude and Cursor, enabling you to:
  • Create, update, archive, restore, and manage pages — the core content units in Architect
  • Manage audiences — targeted versions of pages for different segments and use cases
  • Send messages to draft editors, inspect runs, and work with audience content programmatically
  • Create, search, update, and manage person and account records
  • Pull page, audience, and session analytics from Architect through MCP-driven API calls
  • Create and manage webhook subscriptions for Architect events and delivery workflows

Available tools

Architect’s MCP server exposes two tools: The recommended workflow:
  1. Ask your AI tool to use architect_docs to look up what you want to do.
  2. Then ask it to use architect_api to execute the call.

Authentication

Architect’s MCP uses OAuth 2.0. When you connect your AI tool to Architect’s MCP server, you’ll be redirected to sign in with your Architect account. The MCP server automatically scopes all requests to the tenant tied to the account you authorize.

Available scopes

Supported clients

Setup guides are available for these clients: Other clients that support remote MCP servers can connect with the same server URL. Clients that don’t support remote URLs directly (like Claude Desktop) need a bridge such as mcp-remote — see the setup sections below for exact instructions.

Setup

Connect your AI client to Architect MCP and authorize access to your workspace.

Claude Desktop

1

Open Claude Desktop settings

Open Claude Desktop and go to SettingsDeveloperEdit Config.This opens the claude_desktop_config.json file in your text editor.
2

Add the Architect MCP server

Add the following to the mcpServers section of your config file.If this is your first MCP server (empty or new config file):
If you already have other MCP servers, add "architect" inside the existing mcpServers object. Make sure there’s a comma between each server entry:
Common error: If you see Expected ',' or '}' after property value in JSON, you’re missing a comma between server entries. Every server block must be separated by a comma — but the last entry must not have a trailing comma.
This uses mcp-remote to bridge Claude Desktop’s local transport to Architect’s remote MCP server.
You need Node.js installed for npx to work.
3

Restart Claude Desktop

Close and reopen Claude Desktop for the changes to take effect.
4

Authenticate with Architect

The first time you use an Architect tool, Claude Desktop will open a browser window asking you to sign in to your Architect account. Once signed in, your session is saved and you won’t need to sign in again.

Cursor

1

Open Cursor MCP settings

Open Cursor and go to SettingsMCP.You can also press Cmd+Shift+P (Mac) or Ctrl+Shift+P (Windows/Linux) and search for “MCP: Settings”.
2

Add the Architect MCP server

Click “Add new MCP server” and enter:
  • Name: Architect
  • Type: sse
  • URL: https://mcp.tryarchitect.com/mcp
Alternatively, add it manually to your project-specific or global .cursor/mcp.json file:
3

Authenticate with Architect

When you first use an Architect tool in Cursor’s Agent mode, a browser window will open asking you to sign in to your Architect account. Grant the requested permissions and your session will be saved.
In Cursor, MCP tools are only available in Agent mode (Cmd+I / Ctrl+I), not in regular chat.

Try it

Once connected, ask your AI tool to explore and act on your workspace:
  • “Can you tell me what Architect can do”
  • “Create a person named “Jane Reviewer” with email “[email protected]” in Architect”
  • “Use Architect to pull page analytics for the last 30 days and summarize the top-performing page.”

Security best practices

  • Verify the official endpoint. Always confirm you’re connecting to Architect’s official MCP endpoint: https://mcp.tryarchitect.com/mcp.
  • Understand the access you grant. Connecting to Architect MCP grants the AI tool the same access as your Architect account, scoped to the tenant you authorize. Grant only the scopes the tool needs.
  • Review consent per client. Every new client connection requires explicit OAuth consent — don’t approve authorization prompts you didn’t initiate.
  • Keep human confirmation on. Let your AI tool ask before it runs write actions, so you review each change to pages, records, and subscriptions before it happens.

Troubleshooting

Claude Desktop doesn’t support remote url configs directly. Make sure your config uses "command": "npx" with "args": ["mcp-remote", "https://mcp.tryarchitect.com/mcp"] — not a "url" field. See the config example above.
Your claude_desktop_config.json has a JSON syntax error. This usually means you’re missing a comma between MCP server entries. Make sure each server block is separated by a comma, and that the last entry does not have a trailing comma. Use a JSON validator to check your file.
Confirm the server is configured and your client has been restarted. In Claude Desktop, verify Node.js is installed by running node --version in your terminal. In Cursor, check the MCP server shows a green status indicator in Settings → MCP and that you’re in Agent mode.
Make sure your browser isn’t blocking pop-ups from your AI tool. Try opening https://mcp.tryarchitect.com/mcp directly in your browser to verify the server is reachable, then remove and re-add the server config and authenticate again. If the issue persists, make sure your Architect account has access to at least one workspace.
When you authenticate, make sure to grant all requested permissions. If you need to re-authorize, remove the Architect server from your client, restart it, re-add it, and authenticate again.

Advanced setup

Most clients discover OAuth metadata automatically. Use this only if your client asks for it.