Licensed to be used in conjunction with basebox, only.
// integration
Integrations & API
Applies to
Product: Cloud · Server · Audience: Developer / Integrator
This tab is for everyone who integrates basebox into their own software: developers using the OpenAI-compatible API, and teams connecting MCP servers or their own connectors.
Installation, GPUs, Kubernetes and Helm are in the Operate tab.
Start here: Quickstart
The Quickstart takes you to your first chat completion in a few steps – with curl and with the OpenAI SDK. You need:
- an API key, issued by an administrator of your organisation under Administration → API keys,
- the base URL of your environment,
- the
X-Realmheader with your organisation's realm.
Which values apply for Cloud and Server is in the Quickstart.
The OpenAI-compatible API
basebox provides an OpenAI-compatible API. Tools, libraries and applications written for the OpenAI API work with your basebox environment when you set the base URL to /v1 of your instance and send the X-Realm header. The API is part of every basebox installation and does not need to be switched on.
| Page | Content |
|---|---|
| OpenAI-compatible API | Overview: /v1/chat/completions, /v1/models, Swagger UI |
| User guide | Reference with examples, common mistakes and troubleshooting |
| Streaming | Server-Sent Events, what a client has to handle, reasoning in the stream |
| Error handling | Status codes, messages and recommended client behaviour |
| IDE integration · OpenClaw integration | Connect an editor or coding agent to basebox |
MCP and connectors
| Page | Content |
|---|---|
| MCP overview | What the Model Context Protocol is in basebox; the shipped connectors |
| MCP client | How the assistant discovers, selects and calls tools and frames results |
| Connect an MCP server | Registering an MCP server – in the interface or via Helm |
| Authorization flow | Per-user credentials, access-control gate, security properties |
| Build connectors | Architecture, authentication, read/write access, permissions, examples |
Connecting identity providers, mail, SIEM and business systems is described under Enterprise integration.
Two principles
Permissions stay with the source system. When the assistant accesses an external system through a connector, it does so with the respective person's credentials. basebox neither copies nor extends permissions.
Tool results are data, not instructions. Everything that comes back from a connector or web search is framed as data. Build connectors so that they return structured, bounded results.
Cloud or Server – what changes?
Little. Base URL and X-Realm differ, and on a server the Platform Operator deploys connector services in the cluster. Enterprise integrations apply to both products; network connectivity is established from the customer network for a server and from the basebox environment for the Cloud – see Enterprise integration.
Next step: Quickstart
Need help? Contact support