Skip to main content
Settings → API is where you give things other than you access to a project: your own code, an AI agent, an endpoint that wants to be told when something happens, or an outside service that wants to tell your bot something happened.
This section is for owners and admins. If you’re a workspace member without admin access, it doesn’t appear in your Settings sidebar at all.

What’s on the page

The cards appear in this order, top to bottom:

API Keys

Create a key, tick the permissions it carries, and rotate, revoke, or edit its scopes later. The secret is shown once — copy it there and then.

Connect an AI agent (MCP)

The Paige MCP server address, plus per-client install snippets. Copy-ready for Claude Code, Claude Desktop, and Cursor; the ChatGPT and Codex CLI tabs show the server URL to configure in the client itself.

Connected agents

Every agent connected by sign-in, the projects each one covers, and per-project permissions. Add a project, change permissions, or disconnect.

Inbound hooks

Mint a URL you hand to an outside service — a payment provider, a booking tool, your CRM — so it can call your bot when something happens. Holding the URL is the permission, and it’s redisplayable rather than show-once.

Webhooks

Register an HTTPS endpoint and pick the events to deliver to it. The signing secret is shown once when you create it.

Build an integration with your agent

A ready-made brief you hand to a coding agent so it builds against the live spec instead of guessing. Copy it, download it, or copy the OpenAPI URL.
Only three of the five MCP tabs are copy-paste. ChatGPT’s custom-connector form accepts OAuth, No Authentication, or Mixed Authentication only — there’s no field for a pasted key — and Codex is configured through its own mcp_servers block. Both tabs give you the server URL and a note instead of a snippet. See Connect an AI agent.

Build an integration with your agent

The last card is the fastest way to get an integration written for you. It holds a skill — a ready-made instruction file that tells an AI coding agent how to build against Paige without guessing endpoint shapes. Three buttons:
  • Copy skill — puts the whole skill on your clipboard. Paste it into Claude Code, Codex, Cursor, or whatever you’re building with, and ask for the integration you want.
  • Download .md — saves it as build-a-paige-integration.SKILL.md, so you can commit it alongside your code or drop it into an agent’s skills folder.
  • Copy OpenAPI URL — copies the link to Paige’s live API description.
The skill is served from GET /v1/skill.md, so what you copy is always current. It points the agent at the live OpenAPI document — https://api.paigeme.dev/v1/openapi.json, public and unauthenticated — as its source of truth, and gives it working recipes for sends, templates, broadcasts, contacts, and webhooks. Expand the preview on the card if you want to read it first.
The skill doesn’t include a key, and it can’t make one. Create an API key on the card above first, with only the scopes the integration needs, and give the agent that.

Where to go from here

Each card has a page that explains it properly:
  • Building an integration in your own code? Start at The Paige API, then Authentication, Scopes and API conventions.
  • Connecting Claude, ChatGPT, Codex, or Cursor? Connect an AI agent covers both the sign-in and paste-a-key routes, and how one connection can cover several projects. What a connected agent can do lists every tool it gets.
  • Need an outside service to call your bot? Inbound hooks covers minting the URL, pasting it into a provider’s console, and writing the handler that verifies the sender.
  • Want events pushed to you? Webhooks covers the payload, signature verification, and retries.

Three things that sound alike

Inbound hooks, webhooks, and API keys sit next to each other on this page and are easy to confuse. Ask who starts it: The deciding question is whether you can hand the other side a key. If you can, issue an API key with only the scopes it needs. If you can’t — their console has one field and it says URL — that’s an inbound hook.
Two secrets on this page — an API key and a webhook signing secret — are shown once and never again. Paige stores only what it needs to check them, so nobody can recover one for you. Copy it when the dialog is open.An inbound hook URL is the exception: Show URL brings it back any time. It’s still a secret — holding it is what grants access — so if one is ever exposed, Rotate is how you invalidate it.
Give each integration its own key with only the scopes it needs. When something has to be revoked, you revoke one thing instead of breaking everything at once.