> ## Documentation Index
> Fetch the complete documentation index at: https://docs.connie.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Build on Connie

> The ways developers, scripts and agents can work with Connie: the MCP server (OAuth or API key), webhook triggers, and the API reference.

If you want an AI agent, a script or another system to work with your Connie workspace, there are three ways: the **Connie MCP server**, **Custom Webhook triggers**, and the **[API reference](/api-reference/introduction)**.

<Note>
  **Which should I use?** For anything that runs unattended (a nightly script, a server, an agent), use the **MCP server** with OAuth or an API key: those credentials are built for it and can be revoked on their own. The **API reference** documents the endpoints the Connie app itself uses. They sign in with your browser session, which expires about every hour, so they suit quick scripts, Postman and one-off jobs. See [API authentication](/api-reference/authentication).
</Note>

## Your options

| You want to… | Use | How it signs in |
| - | - | - |
| Let an AI app or agent read and change your workspace (records, lists, workflows, sending, connected apps) | [MCP server](/mcp/overview) at `https://mcp.connie.ai` | **OAuth** (recommended), or a workspace **API key** for apps that can't do OAuth |
| Start a workflow when something happens in another system | [Custom Webhook trigger](/automations/workflows/webhooks) | The webhook's unique URL |
| Call the same endpoints the Connie app uses, from Postman or a quick script | [API reference](/api-reference/introduction) | Your browser session cookie |
| Store a secret (such as another service's API key) for workflow steps to use | [Environment variables](/developers/environment-variables) | Saved in Settings, used as `{{connie.env.KEY}}` |

## The MCP server

MCP (Model Context Protocol) is an open standard for connecting AI apps to services. Connie's MCP server is the programmatic way into Connie. Through it an agent can reach Connie's whole tool catalogue, more than 1,000 tools, with the same credits and permissions as in the app.

* **Address:** `https://mcp.connie.ai`
* **Connection type:** Streamable HTTP. Nothing to install, no local process.
* **Sign-in:**
  * **OAuth (recommended):** you sign in to Connie in a browser and click **Allow access**. The app registers itself automatically, with no client ID or secret. One sign-in reaches every workspace you belong to. Set-up guides: [Claude](/mcp/claude), [ChatGPT](/mcp/chatgpt), [Claude Code](/mcp/claude-code), [Cursor](/mcp/cursor), [VS Code](/mcp/vscode), [other apps](/mcp/other-clients).
  * **Workspace API key:** for apps or scripts that can't sign in with OAuth. Send the header `Authorization: Bearer sk_…`. A key belongs to one workspace and acts with one member's permissions. See [API keys](/developers/api-keys).
* **Tools:** the server lists five tools: **Search tools**, **Get tool**, **Search skills**, **Get skill** and **Use a Connie tool by id**. An agent finds the right tool with Search tools, reads its exact inputs and price with Get tool, then runs it with Use a Connie tool by id. See [Use Connie in your AI app](/mcp/overview).
* **Retries are safe if you send a key:** when running a tool that changes data, sends something or spends credits, pass an idempotency key. If you retry with the same key, Connie returns the original result instead of running the action twice.
* **Billing:** every result from running a tool includes the credits charged. See [Credits](/billing/credits).

To check a new connection works, ask your agent to search Connie's tools for "workspace", or ask "List my Connie workspaces".

## Webhook triggers

A **webhook** is a web address another system calls to tell Connie something happened, for example "a form was submitted" or "a deal closed". Add a **Custom Webhook** trigger to a workflow and Connie gives you a unique URL. Whatever is sent to it becomes variables the workflow can use. See [Webhooks](/automations/workflows/webhooks).

## Good to know

* **Who can use API keys:** owners, admins and members. Guests can't. See [API keys](/developers/api-keys).
* **Credits:** MCP actions use the workspace's credits exactly as in the app. They show in the credit usage log as **MCP**. Actions made with an API key are marked as coming from an API key in [Analytics](/analytics/overview) → Tool activity and in the usage log.
* **Limits:** see [Errors and limits](/developers/errors-and-limits).

## Related

* [Use Connie in your AI app](/mcp/overview)
* [API keys](/developers/api-keys)
* [Environment variables](/developers/environment-variables)
* [Errors and limits](/developers/errors-and-limits)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.