nvidia-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| NVIDIA_API_KEY | Yes | Required. Your `nvapi-...` key. | |
| NVIDIA_TIMEOUT | No | Per-request timeout in seconds. Defaults to 180. | 180 |
| NVIDIA_BASE_URL | No | Point at self-hosted NIM instead. Defaults to https://integrate.api.nvidia.com/v1. | https://integrate.api.nvidia.com/v1 |
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 |
|---|---|
| nv_categoriesA | List every task category this server can route to, with its top models. Start here when you want to know what the NVIDIA catalog is good for. |
| nv_list_modelsA | List models available on the account, tagged by capability. Args: category: optional filter, e.g. "code", "vision", "reasoning", "embedding". refresh: bypass the 1-hour catalog cache. |
| nv_routeA | Recommend the best NVIDIA models for a task, without calling them. Args: task: free-text description of the job, English or Turkish. e.g. "refactor this Python module" / "bu görseldeki tabloyu çıkar". top_k: how many candidate categories to return. |
| nv_healthA | Check that the API key works and report catalog size. |
| nv_probeA | Test which models actually answer on THIS account, and how fast. Being listed by the catalog endpoint does not mean a model is servable — on a free-tier key a large share return 404. Run this to get ground truth. Args: category: probe one category only (e.g. "vision"). Empty probes them all. timeout_seconds: per-model deadline. |
| nv_callA | Call one specific NVIDIA model. Args: model: exact model id, e.g. "mistralai/codestral-22b-instruct-v0.1". prompt: the user message. system: optional system prompt. images: image URLs or local file paths (vision models only). temperature: 0.0-1.0. max_tokens: response cap. |
| nv_autoA | Pick the best model for the task automatically, then run the prompt on it. Args: task: what kind of job this is, e.g. "code review", "görsel analizi". prompt: the actual content to send. system: optional system prompt. images: image URLs or local paths (forces a vision model if given). temperature: 0.0-1.0. max_tokens: response cap. |
| nv_fanoutA | Run one prompt across several models in parallel and return every answer. Use this to cross-check a hard question, compare model behaviour, or build a
quick ensemble. Either pass Args:
prompt: the message every model receives.
models: explicit model ids (max 6). Omit to auto-select from |
| nv_embedA | Embed texts. Returns shape and stats; write vectors to disk with Args: texts: strings to embed. model: embedding model id. Defaults to the best available one. save_to: optional file path to write the vectors as JSON. |
| nv_similarityA | Rank candidate texts by semantic similarity to a query. Args: query: the search string. candidates: texts to rank. model: embedding model id. Defaults to the best available one. |
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 10 tools
Each tool targets a distinct operation: listing models, listing categories, routing, health check, probing, single call, auto call, fanout, embedding, and similarity ranking. The few potentially overlapping pairs (list_models vs categories, route vs auto) are clearly separated by their descriptions, with route explicitly not calling and auto explicitly executing.
Tool names are all prefixed with nv_ and are readable, but they do not follow a consistent pattern: some are verb_noun (nv_list_models), some are bare nouns (nv_categories, nv_health, nv_similarity), and some are bare verbs (nv_route, nv_probe, nv_call, nv_embed). The inconsistent use of underscores (only in nv_list_models) and the mix of verb and noun stems makes the naming scheme unpredictable.
Ten tools is well-scoped for an NVIDIA model inference server: it covers catalog discovery, task routing, account health, availability probing, single and multi-model invocation, and embedding utilities. Each tool serves a distinct purpose, and the count is neither too sparse nor bloated.
The tool set covers the core lifecycle of working with NVIDIA models: discovering models, selecting the right model, verifying access, calling models, and embedding/similarity. Minor gaps exist, such as no dedicated method to fetch detailed model metadata (beyond list filtering) or support for multi-turn chat/streaming, but these are workarounds.