Luma MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| MAX_TOKENS | No | Maximum generation tokens (some models have lower limits) | 8192 |
| MODEL_NAME | No | Override model name (provider-specific) | |
| CUSTOM_PATH | No | API path for custom endpoint | /chat/completions |
| ZHIPU_API_KEY | No | API key for Zhipu (GLM-4.6V) | |
| CUSTOM_API_KEY | No | API key for custom OpenAI-compatible endpoint | |
| MODEL_PROVIDER | No | Model provider: zhipu, siliconflow, qwen, volcengine, hunyuan, custom | zhipu |
| CUSTOM_BASE_URL | No | Base URL for custom endpoint (e.g., https://your-endpoint.com/v1) | |
| HUNYUAN_API_KEY | No | API key for Hunyuan (Hunyuan-Vision-1.5) | |
| CUSTOM_MODEL_NAME | No | Model name for custom endpoint | |
| CUSTOM_TIMEOUT_MS | No | Timeout in milliseconds for custom endpoint | 60000 |
| DASHSCOPE_API_KEY | No | API key for Qwen (DashScope) | |
| BASE_VISION_PROMPT | No | Custom base vision prompt | |
| CUSTOM_AUTH_HEADER | No | Auth header type: bearer, x-api-key, custom | bearer |
| VOLCENGINE_API_KEY | No | API key for Volcengine (Doubao-Seed-1.6) | |
| SILICONFLOW_API_KEY | No | API key for SiliconFlow (DeepSeek-OCR) | |
| CUSTOM_THINKING_MODE | No | Thinking mode: disabled, openai, qwen_extra_body | disabled |
| CUSTOM_AUTH_HEADER_VALUE | No | Custom auth header template (e.g., 'X-API-Key: {{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
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| image_understandA | 图像理解工具（单一入口）：
|
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 1 tool
Only one tool exists, so there is no possibility of misselection between tools. The tool's purpose is clearly defined with explicit call conditions and task types.
The server has a single tool named 'image_understand' in snake_case. With only one tool, there is no naming pattern to conflict; the name is descriptive and consistent with a verb-adjacent structure.
The server exposes exactly one tool. While the tool is comprehensive and well-designed, a single tool feels thin for an MCP server per the calibration guidelines, placing it in the borderline category.
The image_understand tool covers a broad range of use cases (general understanding, OCR, UI analysis, debugging, and description) via its task_type parameter. It provides a complete lifecycle for image understanding, with no obvious missing operations within its stated scope.