Build with AI

If you're using an AI coding assistant, e.g.: Cursor, Claude Code, GitHub Copilot, Claude, you can point it directly at the 7shifts API instead of copying documentation into a chat window.

This page covers what we provide. Start with the MCP Server if you want the short version.

What's available

llms.txt

https://developers.7shifts.com/llms.txt is a machine-readable index of this portal, including our OpenAPI specs. Most AI tools find it automatically. If yours doesn't, give it the URL directly.

Markdown versions of any page

Add .md to any documentation URL to get the raw Markdown:

https://developers.7shifts.com/docs/authentication.md

This is the most reliable way to give an AI tool a specific page as context with no navigation, no HTML, far fewer tokens than a rendered page.

You can also use the Copy Page button at the bottom of any page.

Two things that trip up AI assistants

1. Pin your API version

This one causes real bugs. Your access token or OAuth client is assigned a default API version when it's created, usually whatever was current at the time and not necessarily the latest. Meanwhile, an AI assistant reading our spec may generate code against a different version entirely.

The result is code that looks correct and fails in ways that are hard to trace.

Always set the version explicitly:

curl --request GET --url 'https://api.7shifts.com/v2/companies' \
  --header 'Authorization: Bearer {ACCESS_TOKEN}' \
  --header 'x-api-version: 2026-01-01'

Tell your assistant which version you're targeting at the start of a session. See Versioning on how versioning works.

2. Assistants will suggest OAuth when you don't need it

OAuth is prominent in our documentation, so assistants often reach for it. Most developers don't need it.

Use an access token if you're building against your own company's 7shifts data. OAuth clients are only issued to vetted 7shifts partners. If you have an OAuth client you can still generate an access token on your development account to use when using AI assistants. See Authentication.

If your assistant starts walking you through OAuth client credentials, that's the signal it's gone down the wrong path.

You are responsible for what your agent does

Our MCP server is read-only. It exposes GET endpoints only and nothing an agent does through it can modify or delete your data. That's a deliberate limit, because write operations against 7shifts affect real staff, real schedules, and real payroll, and they can't be undone.

That limit does not extend beyond our MCP server. If you build your own agent, run generated code, or paste an access token into any other AI tool, there's nothing stopping a destructive call. An access token is admin-level and unscoped.

You are responsible for every action taken with your credentials, including actions taken by an AI agent on your behalf.

Practical precautions:

• Review generated code before you run it, particularly anything that writes
• Ask your assistant to show you the request before it makes it
• Never commit an access token to a repository or a shared config file
• Delete access tokens you're no longer using
• Treat anything touching time punches, wages, or schedule publishing as production-critical, because it is

Note that read access is not nothing. An agent can pull wage data, employee records, and time punches into your AI tool's context. Check that's acceptable under your own data handling policies.

Partners with a development account should point agents at that account, not at a customer's.

Rate limits

The 7shifts API allows 10 requests per second per IP address. Agents are considerably chattier than humans, particularly when looping or retrying. If you hit the limit, add backoff. See errors for more details on common errors you may run into.

Getting help

Something inaccurate in our spec or docs that's causing your assistant to generate bad code? That's a documentation bug and we want to hear about it, please post in Discussions.