Let AI agents build and work your forms over MCP
Connect Claude, ChatGPT, Cursor or VS Code with a sign-in link, and let the agent build forms, set up flows and work the inbox.
2 min read
On this page
AI assistants have become good at doing real work, as long as they can reach the tools where the work happens. The Model Context Protocol, MCP for short, is the open standard that gives them that reach. Formpik runs an MCP server, so the assistant you already use can build a form, connect a website, set up a flow or go through yesterday's submissions, the same way you would in the dashboard.
Connect with a link, not a key
The server lives at https://formpik.com/api/mcp. Apps that support MCP sign-in only need that address: they open Formpik's Allow access page, you pick the workspace and how much the agent may do, and the app gets its own token. Nothing to copy, nothing to paste into a config file.
- Claude: add a custom connector with the server address, then click Connect. It works in claude.ai, Claude Desktop and the Claude apps.
- ChatGPT: create a connector with the address and OAuth as the authentication, with Developer mode turned on.
- Cursor and VS Code: add the server to the editor's MCP settings and sign in when it asks.
In Claude Code it is one line, then /mcp and Authenticate:
claude mcp add --transport http formpik https://formpik.com/api/mcpAn agent that runs on its own, with nobody there to sign in, can use an API key in the Authorization header instead.
What you can ask it to do
The tools cover the same ground as the dashboard: forms and their fields, sources and their setup, flows, destinations, submissions, and for admins the team and roles. In practice that means requests like these:
- "Create a job application form with a CV upload, publish it, and give me the link."
- "Add a flow that posts every new demo request to the sales Slack channel and sends the person a confirmation."
- "Show me this week's failed submissions and tell me which step failed."
- "Which source sent the most spam this month?"
The agent answers with what it did, and every change shows up in the dashboard at once, so you can check it, adjust it, or undo it by hand.
It can only do what you allow
When you sign an agent in, you choose its access. Read can only look. Full can also create, change, publish and delete. Admin, which only owners and admins can give, can also manage the team, roles, API keys, settings and billing. On top of that, a connected agent never does more than your own role allows, and nothing can delete the workspace or hand it over. You can see every connected app on the API keys page and disconnect one in a click.
Formpik also tells agents how to behave: ask before deleting anything, before running flows again for a real submission, or before releasing something from spam, because those deliver for real or change a live website.
Agents can send submissions too
An agent that collects information in a conversation, a support bot for example, can post it to a form's endpoint like any website would. It lands in the same inbox, gets the same spam and rule checks, and runs the same flows, so your team handles it exactly like a form someone filled in.
The AI agents page shows the setup for each app, and the MCP guide lists every tool and what it needs.