Connect your AI assistant

AI Studio has one MCP endpoint that any MCP client can connect to. Once connected, your AI assistant can work in your workspace the way you do in the dashboard: find, create and publish agents, run them, read their history and check your credits. This page shows where to find the connection details and what to expect.

Find the connection details#

Open Connect your AI in the AI Studio sidebar. The page is titled Connect your AI assistant and shows:

• MCP server URL, with a Copy URL button.
• A Client picker with setup steps for the client you choose. The page names Claude, ChatGPT and Cursor, and the picker also lists clients such as Claude Code, Codex, Gemini, VS Code, Windsurf and Warp.
• What your agent can do, a list of the tool groups the endpoint offers.
• Connected clients, the AI clients currently signed in to this workspace.

Owners, admins, editors and members can open the page. Guests and maintainers cannot.

Connect a client#

1. Copy the URL

   Click Copy URL on the Connect your AI assistant page.

2. Add it to your client

   Pick your client in the Client picker and follow the steps shown. In Claude Code, for example, you register the URL from the terminal with claude mcp add --transport http. In Claude and ChatGPT you add it as a custom connector.

3. Sign in and choose a workspace

   Your client opens a browser window. Sign in to FlowHunt and pick the workspace on the consent screen.

4. Choose what to allow

   Read-only access is granted by default. Tick the write option if the assistant should create, edit, publish, delete or run things. Separate options cover spending information and chat history.

The connection uses OAuth, so there is no key to copy. A workspace API key also works against the endpoint, but the sign-in route is the one the page walks you through. See API keys.

What the assistant can do#

The tools are grouped by area. In plain terms, an assistant can:

• create, find, edit, publish and delete agents and workflows, and restore earlier versions;
• run an agent, read the result and stop a run, and schedule runs;
• run an agent over a table of inputs and export the results;
• browse templates and start an agent from one;
• manage chatbots, agent grids, knowledge sources and AI Projects;
• read chat history, errors and tool calls, and see credit balances and spending;
• choose workspace models and see which integrations are connected.

It can do only what your role allows in the dashboard. Credits and plan limits apply as well, so with an empty balance it can still read and delete agents but cannot create, edit, publish or run them.

The workspace is fixed#

The connection is bound to the workspace you picked on the consent screen. No tool takes a workspace, so an assistant cannot reach any other workspace. To work in a different one, connect again and pick it.

Running agents costs credits#

When the assistant runs an agent, the run uses the workspace’s credits and can act in connected services, for example by sending an email or posting to Slack. Allow runs only for assistants you trust, and keep your client’s confirmation prompts on. See Credits.

The endpoint also limits how many tool calls a workspace can make per hour. The limit depends on the plan, and calls refused for a missing permission do not count.

Let the assistant read these docs#

The endpoint includes two documentation tools. search_docs finds pages in the AI Studio docs by keyword, and get_docs_page returns a page as Markdown from a path such as /docs/agents/tools/. They touch no workspace data, so an assistant can ask how a feature works before it changes something.

Check and revoke connections#

Connected clients on the same page lists each client with the time it connected. Owners, admins and editors can click Revoke and then Confirm revoke to cut a client off. Members can see the list but not revoke. A removed member or a plan downgrade cuts a connection off on the next call.

Want other clients to use tools you host yourself? See Hosted MCP servers.