io.nodegrove/vram-mcp

Est-ce que io.nodegrove/vram-mcp est sûr ?

Au dernier audit MCP Hub, le serveur MCP « io.nodegrove/vram-mcp » est classé « Non encore testé ». Référencé dans l'annuaire. Seul le pentest comportemental fait foi, et il n'a pas encore été exécuté sur ce serveur. Le verdict est réévalué en continu ; consultez la fiche pour l'état à jour.

Méthode : MCP Hub analyse la description de chaque outil (couche 1) et, pour le verdict fort, exécute le serveur dans une micro-VM isolée avec leurres et réseau bloqué (couche 2). On publie des noms et des compteurs, jamais le contenu brut renvoyé par un outil.

Les 6 outils

• can_i_run≈ 1071 tokens

  Can this GPU run this open-weight LLM? Returns fits, tight or no, the memory split (weights, KV cache, overhead), a decode-speed ceiling, the longest context that fits and, on a no, every change that would make it fit: quantisation, KV cache, context, another card or a smaller model. Model: a name or id from list_models, any Hugging Face repo id, or its architecture. GPU: a name or id from list_gpus, or vram_gb for any other card.

  Paramètres, retour & exemple (10)

  Paramètres

  • gpu · string · optionnel
    A GPU from list_gpus (id or name, e.g. "rtx-4090", "4090" or "M4 Max").
  • model · string · optionnel
    A model from list_models (id or name, e.g. "llama-3.3-70b" or "Llama 3.3 70B"), or any Hugging Face repo id (e.g. "Qwen/Qwen3-8B"), read live from its config.json.
  • quant · string · optionnel · fp16 | q8 | q6 | q5 | q4 | q3
    Weight quantisation: fp16 (FP16 / BF16), q8 (Q8_0), q6 (Q6_K), q5 (Q5_K_M), q4 (Q4_K_M), q3 (Q3_K_M). q4 is the common default.
  • context · integer · optionnel
    Tokens held in context: prompt plus conversation.
  • vram_gb · number · optionnel
    Memory of a card not in list_gpus, GB. For a Mac, its unified memory with apple_silicon: true.
  • kv_cache · string · optionnel · fp16 | q8
    KV cache precision. fp16 is what most runtimes use; q8 halves the cache.
  • architecture · object · optionnel
    A model described by its config.json values instead of a name.
  • apple_silicon · boolean · optionnel
    vram_gb is Apple unified memory; the GPU can use about 75% of it by default.
  • bandwidth_gb_s · number · optionnel
    Memory bandwidth from the maker's spec, GB/s, for a speed ceiling.
  • active_params_b · number · optionnel
    Parameters read per token, billions, for a mixture-of-experts model read from Hugging Face (from its model card). Sets the speed ceiling.

  Exemple d'appel

  {"name": "can_i_run", "arguments": {"gpu": "texte", "model": "texte", "quant": "fp16", "context": 1, "vram_gb": 1, "kv_cache": "fp16"}}

  Exemple de retour

  {"content": [{"type": "text", "text": "…"}]}

• estimate_from_hf_repo≈ 355 tokens

  Reads any Hugging Face model repo's config.json and parameter count and estimates its memory: the attention layout found (standard, sliding-window, hybrid or latent), how much each 1,000 tokens of context costs, and weights + KV cache + overhead at every quantisation. For models nodegrove.io has not reviewed; anything the reader cannot model is listed in warnings.

  Paramètres, retour & exemple (4)

  Paramètres

  • repo · string · requis
    Hugging Face repo id, e.g. "Qwen/Qwen3-8B", or its huggingface.co URL.
  • context · integer · optionnel
    Tokens held in context: prompt plus conversation.
  • kv_cache · string · optionnel · fp16 | q8
    KV cache precision. fp16 is what most runtimes use; q8 halves the cache.
  • active_params_b · number · optionnel
    Parameters read per token, billions, for a mixture-of-experts model read from Hugging Face (from its model card). Sets the speed ceiling.

  Exemple d'appel

  {"name": "estimate_from_hf_repo", "arguments": {"repo": "texte"}}

  Exemple de retour

  {"content": [{"type": "text", "text": "…"}]}

• estimate_vram≈ 869 tokens

  How much memory an LLM needs: weights + KV cache + overhead at each quantisation (or one), at a given context, and the smallest common card class that holds each. Model: a name or id from list_models, any Hugging Face repo id, or its architecture (params_b, layers, kv_heads, head_dim).

  Paramètres, retour & exemple (6)

  Paramètres

  • model · string · optionnel
    A model from list_models (id or name, e.g. "llama-3.3-70b" or "Llama 3.3 70B"), or any Hugging Face repo id (e.g. "Qwen/Qwen3-8B"), read live from its config.json.
  • quant · string · optionnel · fp16 | q8 | q6 | q5 | q4 | q3
    Weight quantisation: fp16 (FP16 / BF16), q8 (Q8_0), q6 (Q6_K), q5 (Q5_K_M), q4 (Q4_K_M), q3 (Q3_K_M). Omit it for all 6.
  • context · integer · optionnel
    Tokens held in context: prompt plus conversation.
  • kv_cache · string · optionnel · fp16 | q8
    KV cache precision. fp16 is what most runtimes use; q8 halves the cache.
  • architecture · object · optionnel
    A model described by its config.json values instead of a name.
  • active_params_b · number · optionnel
    Parameters read per token, billions, for a mixture-of-experts model read from Hugging Face (from its model card). Sets the speed ceiling.

  Exemple d'appel

  {"name": "estimate_vram", "arguments": {"model": "texte", "quant": "fp16", "context": 1, "kv_cache": "fp16", "architecture": {}, "active_params_b": 1}}

  Exemple de retour

  {"content": [{"type": "text", "text": "…"}]}

• list_gpus≈ 135 tokens

  The GPUs and machines nodegrove.io covers: memory, the memory a runtime can use and bandwidth, from the makers' specs, with each one's page.

  Paramètres, retour & exemple (1)

  Paramètres

  • search · string · optionnel
    Words to filter by, e.g. "qwen" or "24 GB".

  Exemple d'appel

  {"name": "list_gpus", "arguments": {"search": "texte"}}

  Exemple de retour

  {"content": [{"type": "text", "text": "…"}]}

• list_models≈ 152 tokens

  The open-weight LLMs nodegrove.io has verified against their config.json (data version 2026-10-16): id, size, attention design, native context, licence, memory at Q4 with 8k context and each model's page.

  Paramètres, retour & exemple (1)

  Paramètres

  • search · string · optionnel
    Words to filter by, e.g. "qwen" or "24 GB".

  Exemple d'appel

  {"name": "list_models", "arguments": {"search": "texte"}}

  Exemple de retour

  {"content": [{"type": "text", "text": "…"}]}

• what_fits≈ 418 tokens

  Which open-weight LLMs fit this GPU: every model in list_models checked at one quantisation and context, with a recommended everyday model (the biggest class that fits with room for context at conversational speed), the largest that fits, the best at Q8 and the first out of reach. GPU: a name or id from list_gpus, or vram_gb for any other card.

  Paramètres, retour & exemple (6)

  Paramètres

  • gpu · string · optionnel
    A GPU from list_gpus (id or name, e.g. "rtx-4090", "4090" or "M4 Max").
  • quant · string · optionnel · fp16 | q8 | q6 | q5 | q4 | q3
    Weight quantisation: fp16 (FP16 / BF16), q8 (Q8_0), q6 (Q6_K), q5 (Q5_K_M), q4 (Q4_K_M), q3 (Q3_K_M). q4 is the common default.
  • context · integer · optionnel
    Tokens held in context: prompt plus conversation.
  • vram_gb · number · optionnel
    Memory of a card not in list_gpus, GB. For a Mac, its unified memory with apple_silicon: true.
  • apple_silicon · boolean · optionnel
    vram_gb is Apple unified memory; the GPU can use about 75% of it by default.
  • bandwidth_gb_s · number · optionnel
    Memory bandwidth from the maker's spec, GB/s, for a speed ceiling.

  Exemple d'appel

  {"name": "what_fits", "arguments": {"gpu": "texte", "quant": "fp16", "context": 1, "vram_gb": 1, "apple_silicon": true, "bandwidth_gb_s": 1}}

  Exemple de retour

  {"content": [{"type": "text", "text": "…"}]}

C'est votre serveur ?

Revendiquez-le sur l'espace créateur (preuve GitHub, domaine ou npm) pour afficher le badge « propriétaire vérifié » et lancer un pentest comportemental à la demande afin d'obtenir le badge « Vérifié ».

Créer un compte / revendiquer →

Source

https://mcp.nodegrove.io/mcp