Docs

MCP server

Connect Claude, ChatGPT or Cursor to Callables, so an assistant can deploy, run and fix callables for you without git.

The Callables MCP server lets an AI assistant deploy, run and fix callables for you. Ask Claude, ChatGPT or Cursor for an automation in plain words. It writes the code, deploys it to your workspace, runs it and fixes what breaks, and you get a link to a working callable your team can run. You don't need git, a terminal or the CLI.

The server's address is:

https://api.callables.app/mcp

The first time you use it, your assistant opens a Callables page in your browser. Sign in, pick the workspace, and approve what it may do. It then acts as you in that workspace, and can't do anything your role doesn't allow.

Connect your assistant

Claude Code

claude mcp add --transport http callables https://api.callables.app/mcp

Then run /mcp in Claude Code and choose Callables to sign in.

Claude (web and desktop)

In Settings, open Connectors, choose Add custom connector, and paste the address.

Cursor

Add this to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for every project:

{
  "mcpServers": {
    "callables": { "url": "https://api.callables.app/mcp" }
  }
}

ChatGPT and other assistants

Any assistant that supports remote MCP servers with OAuth can connect: add a custom connector or app with the address above.

What it can do

ToolWhat it does
get_guideReads the guide to writing callables, so the assistant writes them correctly
list_projectsLists your projects and their latest deploys
deploy_filesCreates a project from files, or replaces an uploaded project's files, and starts a build
redeployRebuilds a project from its newest files, or a GitHub project from its latest commit
get_deployShows a build's status, errors and log, and the callables it published
list_callablesLists the callables you can run, their inputs, and any missing secrets or connections, with links to set them
run_callableRuns a callable as you and waits for the result
get_runShows a run's result, error and logs
list_runsLists recent runs, yours and every run (scheduled ones too) of callables you manage, filtered by status or time
get_sourceGets an uploaded project's files, so a new chat can change them

Assistants ask before deploying or running a callable, because those change things.

Ask "how are my automations doing?" and the assistant checks recent runs with list_runs, reads the logs of a failed one, fixes the code and deploys again.

Permissions

The approval page lists what the assistant asks for:

  • Read: see your projects, deploys, callables and runs.
  • Deploy: create projects and deploy code to them.
  • Run: run callables as you.

It can't set, change or read secrets. You set them in the web app.

It also only gets what your role allows. A member can run callables shared with them but can't deploy.

Whoever manages a callable can also stop assistants from running it: on the callable's page in the builder console, turn off Assistants can run this function. Assistants still see it, and people can still run it themselves in the web app. It's on by default.

To disconnect an assistant, open Connections in the web app and choose Disconnect under Connected apps. Its access stops at once.

Uploaded projects and GitHub projects

A project the assistant creates with deploy_files holds its files in Callables instead of a git repo. Each deploy replaces all of its text files, and Callables keeps the newest set so the assistant can fetch them later with get_source. Binary files, such as images or data files from a .zip you uploaded, stay in the project when the assistant deploys, because it can only send text. It can delete one by name.

You can also upload a new .zip of an uploaded project yourself, from its page.

If your code lives in GitHub, connect the repo in the web app as usual. Every push deploys, and the assistant can still follow builds, run callables and check on runs. It won't upload files over a GitHub project: it changes the code in the repo instead.

Secrets and connections

The assistant never handles your API keys. When a callable needs a secret, list_callables tells it which ones are missing and gives it a link to the project's secrets. You add the value there, in the web app, and it's never pasted into the chat. The assistant also skips .env files when it deploys.

Connections, where a callable acts on someone's own Google or Slack account, need that person to sign in from the web app. The assistant gives you the link.

Safety

An assistant that can deploy and run can do anything code in your project can do. Keep these in mind:

  • Assistants ask before deploying or running. Read what they're about to run, especially after they've read emails, web pages or other text someone else wrote: that text can contain instructions meant to trick the assistant.
  • Secrets belong to one project, so code an assistant deploys in a new project can't read your other projects' keys.
  • Disconnect assistants you no longer use.

Example

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, writes the function, deploys it, runs it with a sample URL, and replies with a link to the callable. From there, anyone in your workspace you share it with can run it from a form.