Luma MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TOP_P | No | Nucleus sampling threshold. | 0.95 |
| LUMA_DEBUG | No | `1`/`true` enables the same meta output as INCLUDE_META. | |
| MAX_TOKENS | No | Maximum number of generated tokens. Note: for SiliconFlow (DeepSeek-OCR) the client hard-caps this at 4096. | 8192 |
| MODEL_NAME | No | Model name override; when unset the provider's default model is used (e.g. `doubao-seed-1-6-vision-250815`, `hy-vision-2.0-instruct`). | 自动选择 |
| MULTI_CROP | No | Multi-crop for large images; set to `false` to disable. | true |
| CUSTOM_PATH | No | Custom provider API path. | /chat/completions |
| TEMPERATURE | No | Sampling temperature. | 0.7 |
| INCLUDE_META | No | When `true`, appends preprocessing/API timing meta (`luma_meta`) to the tool result. | false |
| MCP_HTTP_HOST | No | HTTP mode listen address (must be `0.0.0.0` inside Docker). | 0.0.0.0 |
| MCP_HTTP_PORT | No | HTTP mode listen port. | 3000 |
| MCP_TRANSPORT | No | Transport mode: `stdio` (default) or `http` (Streamable HTTP). | stdio |
| ZHIPU_API_KEY | No | API key for the Zhipu provider (default model: `glm-4.6v`). | |
| CUSTOM_API_KEY | No | API key for the custom OpenAI-compatible endpoint (MODEL_PROVIDER=custom). | |
| MCP_HTTP_TOKEN | No | Bearer token for HTTP mode; strongly recommended when sharing the server on a LAN, since otherwise anyone reaching the port can call `image_understand`. | |
| MODEL_PROVIDER | No | Model provider: `zhipu`, `siliconflow`, `qwen`, `volcengine`, `hunyuan`, or `custom`. | zhipu |
| CUSTOM_BASE_URL | No | OpenAI-compatible API base URL for the custom provider (e.g. `https://your-endpoint.com/v1`, `https://tokenhub.tencentmaas.com/v1`). | https://api.minimaxi.com/v1 |
| ENABLE_THINKING | No | Thinking mode; set to `false` to disable. | true |
| HUNYUAN_API_KEY | No | API key for the Hunyuan provider (default model: `hunyuan-t1-vision-20250916`; newer HY-Vision models are served via Tencent Cloud TokenHub and should be used with the `custom` provider). | |
| CUSTOM_MODEL_NAME | No | Model name for the custom provider. | MiniMax-M3 |
| CUSTOM_TIMEOUT_MS | No | Custom provider request timeout in milliseconds. | 60000 |
| DASHSCOPE_API_KEY | No | API key for the Qwen (DashScope) provider (default model: `qwen3-vl-flash`). | |
| BASE_VISION_PROMPT | No | Custom base vision prompt (set to an empty string to disable). | |
| CUSTOM_AUTH_HEADER | No | Custom provider auth header style: `bearer` / `x-api-key` / `custom`. | bearer |
| VOLCENGINE_API_KEY | No | API key for the Volcengine provider (default model: `doubao-seed-1-6-flash-250828`). | |
| SILICONFLOW_API_KEY | No | API key for the SiliconFlow provider (default model: `deepseek-ai/DeepSeek-OCR`). | |
| CUSTOM_THINKING_MODE | No | Custom provider thinking mode: `disabled` / `openai` / `qwen_extra_body`. | disabled |
| MULTI_CROP_MAX_TILES | No | Maximum number of crop tiles (including the original image), 1–16. | 5 |
| CUSTOM_AUTH_HEADER_VALUE | No | Custom Header template for the custom provider when CUSTOM_AUTH_HEADER=custom, e.g. `X-API-Key: {{key}}` (`{{key}}` is replaced with the API key). |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Server capabilities have not been inspected yet.
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
No tools | |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
This server cannot be deployed
Maintenance
ActivityActive
ResponsivenessNo issues