Connect your tool

One key — mxn-… from your account. Only the address changes, and it depends on the client’s protocol, not on the model:

The client speaks	Base URL	Who
OpenAI (Chat Completions, Responses)	https://api.mixen.ai/v1	Cursor, Cline, Codex CLI, OpenCode, OpenAI SDK, n8n, Dify
Anthropic (Messages)	https://api.mixen.ai — no /v1	Claude Code, Claude Desktop, Anthropic SDK, OpenClaw
Gemini (generateContent)	https://api.mixen.ai/v1beta	Gemini CLI, GoogleGenAI SDK
MCP	https://api.mixen.ai/v1/mcp	Any MCP-capable agent

The most common mistake is taking the address for the wrong protocol: Claude Code appends /v1/messages itself, so /v1 in the base URL doubles the path. Every tile below already carries the right one.

Terminal agents

Claude Code

Anthropicapi.mixen.ai

Wrapper script or settings.json; tools, streaming, context caching and thinking all work.

Connect in 2 minutes →

Codex CLI

OpenAIapi.mixen.ai/v1

A provider in config.toml, Responses API with tool calls.

Connect in 2 minutes →

OpenCode

OpenAIapi.mixen.ai/v1

A custom provider in opencode.json, agent modes and your own model set.

Connect in 2 minutes →

Gemini CLI

Geminiapi.mixen.ai/v1beta

Two environment variables — and the whole Mixen catalog inside Gemini CLI.

Connect in 2 minutes →

Qwen Code

OpenAIapi.mixen.ai/v1

A provider in ~/.qwen/settings.json with the openai protocol.

Connect in 2 minutes →

ZCode

OpenAI · Anthropicapi.mixen.ai/v1

A custom provider with OpenAI- and Anthropic-compatible routes.

Connect in 2 minutes →

IDEs and editors

Cursor

OpenAIapi.mixen.ai/v1

Override OpenAI Base URL, your key, a model per task.

Connect in 2 minutes →

Cline

OpenAIapi.mixen.ai/v1

The OpenAI Compatible provider, Plan and Act modes.

Connect in 2 minutes →

Roo Code

OpenAIapi.mixen.ai/v1

The same provider profile, Code / Architect / Ask modes.

Connect in 2 minutes →

Kilo Code

OpenAIapi.mixen.ai/v1

OpenAI Compatible in the VS Code extension settings.

Connect in 2 minutes →

Continue

OpenAIapi.mixen.ai/v1

The openai provider in config.yaml, chat and autocomplete.

Connect in 2 minutes →

Apps and autonomous agents

Claude Desktop

Anthropicapi.mixen.ai

The Code tab through a third-party inference gateway in Developer Settings.

Connect in 2 minutes →

OpenClaw

Anthropicapi.mixen.ai

A custom anthropic-messages provider in openclaw.json.

Connect in 2 minutes →

Hermes Agent

OpenAIapi.mixen.ai/v1

model.base_url in config.yaml, the openai_chat transport.

Connect in 2 minutes →

Mixen MCP server

MCPapi.mixen.ai/v1/mcp

The agent gets chat, image generation and balance as tools — no integration code.

Connect in 2 minutes →

SDKs and code

OpenAI SDK

OpenAIapi.mixen.ai/v1

Python and JavaScript: change base_url and the key, nothing else.

First request →

Anthropic SDK

Anthropicapi.mixen.ai

Messages API with tools, streaming and count_tokens.

Python example →

LangChain

OpenAIapi.mixen.ai/v1

ChatOpenAI with base_url — chains, agents, structured output.

Connect in 2 minutes →

Vercel AI SDK

OpenAIapi.mixen.ai/v1

createOpenAI({ baseURL }) — streaming and tools in Next.js.

Connect in 2 minutes →

Chat UIs and no-code

Open WebUI

OpenAIapi.mixen.ai/v1

Your own ChatGPT-style UI with the whole Mixen catalog.

Connect in 2 minutes →

n8n

OpenAIapi.mixen.ai/v1

OpenAI credentials with a Base URL — automations without code.

Connect in 2 minutes →

Dify

OpenAIapi.mixen.ai/v1

The OpenAI-API-compatible provider for Dify apps and agents.

Connect in 2 minutes →

Your client is not here? If it lets you set a custom base URL, it connects with one of the four rows above. What is not supported (Assistants API, Batch, fine-tuning) and how each protocol names a tool call — in the integrations overview.