HelloLeo MCP is the toolset that lets Leo act on your HelloLeo account itself, not just on the code of the project in front of it: your other projects, your AI agents, their automations, and the integrations you have already connected.
It works in two directions, and it helps to keep them apart.
Inside HelloLeo it is on by default, on every new project and every agent. Nothing to enable, nothing to paste. It is what turns the chat into a single place to work from: ask for an agent and Leo creates one, ask it to reuse a component from another project and it goes and reads it.
From outside HelloLeo the same toolset is a remote MCP server. Connect it to Claude, Claude Code, Cursor or any MCP client and that assistant works on your HelloLeo account, with your permissions.
What you can ask for in the chat
You do not need to know which menu holds what. Describe the outcome and Leo picks the right thing.
Work that runs on its own
“Alert me every morning when a customer invoice is more than 15 days overdue.”That is an agent with a schedule, not an app. Leo creates the agent, connects the integration it needs, and hands it back to you.
Something your team opens and uses
“Build a page where my sales team can ask about open orders.”Here an interface earns its place, so Leo builds the app and wires it to the agent.
Across your other projects
“Copy the invoice table from my Client Portal into this project.”Leo reads the other project and brings the component over.
Choosing what Leo may do
The toolset is grouped, and each group has a switch. You will find them in Project Settings → HelloLeo AI for a project, and in the Permissions tab for an agent.
Everything is on by default, with one exception: Read and edit your other projects starts off in the builder, because the chat already has file tools for the project you are in and two near-identical toolsets only confuse things. It is on for agents, which have no file tools of their own.
Turning a group off removes those tools entirely, from the next message onwards — including from an external client connected to your account.
Use it from your AI client
Connecting an external AI client requires a plan that includes HelloLeo MCP (Leo Business and above). Inside HelloLeo the toolset is available on every plan.
Claude (web and desktop)
Add the connector
Settings → Connectors → Add custom connector, paste the URL above.
Connect
Click Connect. Your browser opens the HelloLeo consent screen, approve it.
Use it in a conversation
Enable the connector in the chat, then try “List my HelloLeo agents.”
Claude Code
The first tool call triggers the browser sign-in.
Cursor and other MCP clients
Add a remote MCP server (streamable HTTP transport) pointing at the URL above. Any client that supports remote MCP servers with OAuth will work the same way.
What it can and cannot reach
Every call is checked against your own permissions, on our side, not the client’s. An assistant connected to your account sees exactly the projects you see: the ones you own, plus those shared through a workspace you belong to. Anything else answers “not found”.
A few things are deliberately out of reach, whatever the assistant asks for: environment files and secrets, the internal git directory, and HelloLeo’s own internal files. Requests are also rate limited, so a runaway loop cannot hammer your projects.
Troubleshooting
The client says the server is unauthorized. Sign in to app.helloleo.dev in the same browser, then reconnect. The consent screen needs an active HelloLeo session.
Connecting an external client fails and your plan is below Business. External clients are gated by plan. Inside HelloLeo the toolset works regardless, or upgrade to connect your own assistant.
Leo says it cannot create an agent. Check the switches above: Create and change agents may be off for that project or agent.
A project created before this shipped has no toolset. Projects created earlier keep whatever they had. Open Project Settings → HelloLeo AI to see the state for that project.
A newly created project has no files. A blank project starts empty: ask the assistant to write its first file, then to deploy it. Preview URLs only appear once the sandbox has booted.
Analytics come back empty. A project only has analytics once it has been previewed or published, and only page-level traffic is measured. Individual clicks are not tracked. 