vLLM MCP Server
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| VLLM_MODEL | No | Default model to use | |
| VLLM_API_KEY | No | API key (if required) | |
| VLLM_BASE_URL | No | vLLM server URL | http://localhost:8000 |
| VLLM_HF_TOKEN | No | HuggingFace token for gated models (e.g., Llama) | |
| VLLM_DOCKER_IMAGE | No | Container image (GPU mode) | vllm/vllm-openai:latest |
| VLLM_CONTAINER_NAME | No | Container name | vllm-server |
| VLLM_DEFAULT_TIMEOUT | No | Request timeout (seconds) | 60.0 |
| VLLM_DOCKER_IMAGE_CPU | No | Container image (CPU mode) | quay.io/rh_ee_micyang/vllm-cpu:v0.11.0 |
| VLLM_CONTAINER_RUNTIME | No | Container runtime (podman, docker, or auto) | |
| VLLM_DEFAULT_MAX_TOKENS | No | Default max tokens | 1024 |
| VLLM_DOCKER_IMAGE_MACOS | No | Container image (macOS) | quay.io/rh_ee_micyang/vllm-mac:v0.11.0 |
| VLLM_DEFAULT_TEMPERATURE | No | Default temperature | 0.7 |
| VLLM_GPU_MEMORY_UTILIZATION | No | GPU memory fraction | 0.9 |
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": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| vllm_chatC | Send a chat message to the vLLM server. Supports multi-turn conversations. |
| vllm_completeB | Generate text completion using vLLM. Good for code completion and text generation. |
| list_modelsB | List all available models on the vLLM server |
| get_model_infoC | Get detailed information about a specific model |
| vllm_statusB | Check the health and status of the vLLM server |
| start_vllmA | Start a vLLM server in a Docker container. Automatically detects platform (Linux/macOS/Windows) and GPU availability. |
| stop_vllmB | Stop a running vLLM Docker container |
| restart_vllmC | Restart a vLLM Docker container |
| list_vllm_containersC | List all vLLM Docker containers |
| get_vllm_logsB | Get logs from a vLLM container to check loading progress or errors |
| get_platform_statusB | Get platform information including Docker and GPU availability |
| run_benchmarkB | Run a performance benchmark against the vLLM server using GuideLLM |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| coding_assistant | A helpful coding assistant that writes clean, efficient code |
| code_reviewer | Reviews code for bugs, security issues, and improvements |
| technical_writer | Creates clear technical documentation |
| debugger | Helps identify and fix bugs in code |
| architect | Designs software systems and architectures |
| data_analyst | Analyzes data and creates insights |
| ml_engineer | Develops machine learning models and pipelines |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| vLLM Server Status | Current status and health of the vLLM server |
| vLLM Performance Metrics | Performance metrics from the vLLM server |
| vLLM MCP Configuration | Current configuration settings |
| Platform Information | Platform, Docker, and GPU status information |
TDQS
Scored across 12 tools
Most tools have distinct purposes, but there is some potential confusion between vllm_chat and vllm_complete for text generation tasks, and between get_model_info and list_models for model information. Descriptions help clarify, but the overlap could cause misselection in ambiguous scenarios.
The naming is mixed with inconsistent patterns: some use verb_noun (e.g., get_model_info, list_models), others use noun_verb (e.g., vllm_chat, vllm_complete), and there are variations like run_benchmark. While readable, the lack of a uniform convention reduces predictability.
With 12 tools, the count is well-scoped for managing a vLLM server, covering model operations, container management, and inference tasks. Each tool appears to serve a specific role without unnecessary duplication, fitting the server's purpose effectively.
The tool set provides comprehensive coverage for vLLM server management, including deployment (start/stop/restart), monitoring (status/logs), and usage (chat/completion). A minor gap is the lack of tools for model updates or configuration changes, but core workflows are well-supported.