Introducing the Callables MCP server (beta)
Today we're opening the beta of the Callables MCP server. Connect it to Claude, ChatGPT, Cursor or any assistant that supports remote MCP servers, and you can ask for an automation in plain words. The assistant writes the Python, deploys it to your workspace, runs it, reads the logs, fixes what broke and deploys again. You get a link to a callable your team can run from a form.
You don't need git, a terminal or the CLI. If you can describe the job, you can ship it.
Why we built it
Callables started as the fastest way to turn a Python function into a tool your team can use: type hints become a form, the return value becomes a table or a file, and roles decide who can run what. The people who got the most out of it already wrote code, or worked next to someone who did.
Meanwhile, a lot of people now build software by talking to an assistant. They can get working Python out of Claude in a minute. What they can't easily do is put that script somewhere safe and shared: with secrets kept out of the chat, a schedule, run history, and a form for the teammate who'll never open a terminal. That last step is Callables' whole job. The MCP server lets the assistant take it for you.
What it looks like
Add the server:
https://api.callables.app/mcpIn Claude, that's Settings > Connectors > Add custom connector. In Claude Code it's one command:
claude mcp add --transport http callables https://api.callables.app/mcpThe first time, a Callables page opens in your browser. Sign in, pick the workspace and approve what the assistant may do. Then ask:
Make a callable that takes a list of URLs and returns each page's title and status code as a table.
The assistant reads the guide to writing callables, writes the function, deploys it, runs it with a sample URL, and replies with a link. Anyone you share it with can run it from a form.
It keeps working after the first deploy. Ask "how are my automations doing?" and the assistant checks recent runs, opens the logs of a failed one, fixes the code and deploys the fix.
What the assistant can do
| Tool | What it does |
|---|---|
get_guide | Reads the guide to writing callables, so its code is right the first time |
list_projects | Lists your projects and their latest deploys |
deploy_files | Creates a project from files, or updates one, and starts a build |
redeploy | Rebuilds a project from its newest files or latest commit |
get_deploy | Shows a build's status, errors and log |
list_callables | Lists what you can run, its inputs, and any missing secrets or connections |
run_callable | Runs a callable as you and waits for the result |
get_run / list_runs | Reads a run's result and logs, or recent runs, scheduled ones included |
get_source | Fetches a project's files, so a new chat can pick up where the last one stopped |
It acts as you, and only as you
Connecting an assistant doesn't create a new kind of user. The assistant signs in with OAuth and acts as you in one workspace, with three permissions you approve: read, deploy and run. It can't do anything your role doesn't allow. A member who can run shared callables can let an assistant run them too, but not deploy.
A few rules matter more with an assistant in the loop:
- It never touches secrets. It can't set, change or read them. When a callable needs an API key, the assistant tells you which one is missing and links you to the project's secrets page. You paste the value there, not into the chat.
- Each callable's manager decides whether assistants can run it. Every callable has an Assistants can run this function switch, on by default. Turn it off for something sensitive, like issuing refunds, and assistants still see it, but only people can run it.
- Assistants ask first. Deploying and running are marked as actions that change things, so your assistant confirms before each one.
- You can disconnect at any time. Connections > Connected apps in the web app shows every assistant you've connected. Disconnecting one cuts off its access at once.
Projects without git
Most Callables projects are GitHub repos, and those keep working as before. Every push deploys, and the assistant can follow builds and runs. For everyone else there are now uploaded projects: the assistant sends the files, Callables keeps the latest version, and a new chat can fetch them to keep going. You can also upload a .zip of a project folder yourself from New project.
Beta notes
The server is built on the open MCP standard, so any assistant that supports remote servers with OAuth can connect. We've tested it with Claude, Claude Code, ChatGPT and Cursor. It's a beta: expect rough edges, and expect it to change quickly.
Tell us what you build, and what got in the way.
Read the docs to connect your assistant.
Turn your scripts into tools your team can run
Add @deploy to a Python function and push. Your team gets a form, schedules and run history.
Keep reading
Build automations with Claude Code, Cursor and other AI agents
One command teaches your coding agent how to write Callables functions, check them and run them locally. Then you push.
GuidesControl who can run your automations
Roles for your team, sharing for each function, and run history that only the right people see. How access works in Callables.
GuidesHow to automatically extract invoices from Gmail and Outlook into a spreadsheet
Automate invoice processing with Python and Claude: find every invoice attached to your emails, pull the vendor, date, number and total into a CSV, and download all the PDFs as one zip.
