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

# Overview

BundleUp provides MCP (Model Context Protocol) servers that let AI agents interact with external systems through your integrations.

Instead of writing custom tool definitions for every provider, you can expose integrations like Slack, GitHub, Jira, or Google Drive directly to AI models through a structured MCP interface.

BundleUp offers two MCP server types, each designed for a different use case.

## **Authentication**

Every MCP request must include:

* **API key** in the Authorization header
* **Connection ID** in the BU-Connection-Id header

```
Authorization: Bearer <YOUR_API_KEY>
BU-Connection-Id: <CONNECTION_ID>
```

The API key identifies your workspace, and the connection ID selects the external account the tools run against. If either header is missing or invalid, the MCP server returns an error before any upstream API is contacted.

## **Errors**

Common responses include:

* **401 Unauthorized** when the API key is missing or invalid.
* **400 Bad Request** with the `connection_invalid` error code when the connection ID is missing or invalid.
* **429 Too Many Requests** when the per-connection rate limit is exceeded. A `Retry-After: 60` header indicates when you can try again.

Upstream API errors are returned inside the MCP tool response so clients can surface them directly to users.