Set up Typeform MCP Server

The Typeform MCP Server connects your AI agent to your Typeform account: create forms, analyze responses, and set up automations — directly from your AI client. Start with read tools and enable write tools only when needed.

Establish the connection

For Claude Code, a single command is enough: claude mcp add --transport http typeform https://api.typeform.com/mcp. For Claude Desktop or the web interface, go to Settings → Connectors, search for Typeform, and click Connect. For other clients (Cursor, VS Code, Codex), add the server in your MCP configuration file with the URL https://api.typeform.com/mcp and HTTP transport. On first connection, a browser window opens for the Typeform OAuth login. Important: SSE transport is not supported by the server — make sure your client uses Streamable HTTP.

First call: get the account ID

After connecting, call the accounts-list_accounts tool first. Almost all other tools require the returned account_id as a parameter. Without this step, subsequent requests will fail with an authorization error. This is a characteristic of the Typeform API model, not a server bug.

Use tools selectively

The server provides four tool groups according to the provider: forms-* (create and edit forms), automations-* (build workflows), contacts-* (contact management), and insights-* (response analytics). For pure analytics tasks, the insights-* tools are sufficient. Write tools in forms-* and automations-* can modify forms and workflows — only enable these when the agent is actually supposed to make changes.

Avoid prompt injection

When the agent analyzes form responses, user input enters the language model's context. This input may contain instructions that lead the agent to unintended actions (prompt injection). Separate read and write operations: analyze responses in a separate session without write tools enabled, or manually confirm write operations before the agent executes them.