LLM Configurator
Read-only: which local LLMs a GPU or Mac can run - VRAM fit, tokens/sec, model specs.
我该使用它吗
质量与安全性
A
基于对工具定义和协议合规性的自动分析。
上下文开销
~4,728token 数（工具定义）
~9.3 KB典型响应大小
对注意力有显著影响（占 128k 上下文窗口的 3.69%）
这是每次将服务器的工具加载到模型上下文窗口时所消耗的大致 token 数。数值越高，可用于其他任务的注意力就越少。
安装
一键安装
将以下内容添加到你的 `claude_desktop_config.json` 文件中：
{
"mcpServers": {
"llmconfigurator": {
"url": "https://elgadmisijrfowuujcmu.supabase.co/functions/v1/mcp"
}
}
}远程端点
https://elgadmisijrfowuujcmu.supabase.co/functions/v1/mcpstreamable-http它能做什么
工具清单
工具（5）
🟢 只读🟡 写入🔴 删除⚪ 未知
🟢search_catalog(query, kind, limit)
Use this first when the user names a GPU, Mac, mini PC or language model and you need its id for the other tools. Matches names and common aliases in a fixed catalogue of local-LLM hardware and models. It does not return specs, fit or speed, and it does not search the web.
输入模式
{
"type": "object",
"properties": {
"query": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[^\\u0000-\\u001f\\u007f]*$",
"description": "Hardware or model name as the user wrote it, e.g. \"4090\", \"m4 max 64\", \"llama 3.3 70b\"."
},
"kind": {
"type": "string",
"enum": [
"hardware",
"model"
],
"description": "Search hardware or models. Required."
},
"limit": {
"default": 5,
"description": "Maximum matches, 1 to 10. Default 5.",
"type": "integer",
"minimum": 1,
"maximum": 10
}
},
"required": [
"query",
"kind"
]
}输出模式
{
"type": "object",
"properties": {
"data_as_of": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$",
"description": "Date the bundled catalogue was last updated. Not the time of this call."
},
"summary": {
"type": "string",
"maxLength": 600,
"description": "One or two plain-language sentences stating the answer."
},
"kind": {
"type": "string",
"enum": [
"hardware",
"model"
]
},
"matches": {
"type": "array",
"items": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$",
"description": "Pass this id to the other tools."
},
"name": {
"type": "string",
"maxLength": 120
},
"kind": {
"type": "string",
"enum": [
"hardware",
"model"
]
},
"detail": {
"type": "string",
"maxLength": 160,
"description": "A short factual line, e.g. \"24 GB, 1008 GB/s\" or \"8B dense\"."
},
"url": {
"type": "string",
"pattern": "^https:\\/\\/llmconfigurator\\.com\\/en\\/"
}
},
"required": [
"id",
"name",
"kind",
"detail",
"url"
],
"additionalProperties": false
}
}
},
"required": [
"data_as_of",
"summary",
"kind",
"matches"
],
"additionalProperties": false
}🟢check_hardware_fit(hardware_id, model_id, quantisation, context_length)
Use this when the user asks whether one specific model runs on one specific machine, at what memory cost, and roughly how fast. Needs a hardware id and a model id from search_catalog. Returns a fit verdict, the memory arithmetic, and a decode-speed range with a confidence label. Memory is sized from all parameters, speed from active parameters. It does not recommend purchases and does not run anything.
输入模式
{
"type": "object",
"properties": {
"hardware_id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$",
"description": "Hardware id from search_catalog."
},
"model_id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$",
"description": "Model id from search_catalog."
},
"quantisation": {
"default": "Q4_K_M",
"description": "Weight quantisation. One of Q2_K, Q3_K_M, Q4_K_M, Q5_K_M, Q6_K, Q8_0, F16. Default Q4_K_M.",
"type": "string",
"enum": [
"Q2_K",
"Q3_K_M",
"Q4_K_M",
"Q5_K_M",
"Q6_K",
"Q8_0",
"F16"
]
},
"context_length": {
"default": 4096,
"description": "Context window in tokens that the KV cache is sized for. 512 to 262144. Default 4096.",
"type": "integer",
"minimum": 512,
"maximum": 262144
}
},
"required": [
"hardware_id",
"model_id"
]
}输出模式
{
"type": "object",
"properties": {
"data_as_of": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$",
"description": "Date the bundled catalogue was last updated. Not the time of this call."
},
"summary": {
"type": "string",
"maxLength": 600,
"description": "One or two plain-language sentences stating the answer."
},
"verdict": {
"type": "string",
"enum": [
"fits_comfortably",
"fits_tight",
"needs_offload",
"does_not_fit"
]
},
"quantisation": {
"type": "string",
"enum": [
"Q2_K",
"Q3_K_M",
"Q4_K_M",
"Q5_K_M",
"Q6_K",
"Q8_0",
"F16"
]
},
"context_length": {
"type": "integer",
"minimum": -9007199254740991,
"maximum": 9007199254740991
},
"model": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$"
},
"name": {
"type": "string",
"maxLength": 120
},
"total_params_b": {
"type": "number",
"exclusiveMinimum": 0,
"description": "Total parameters in billions. Sets memory residency."
},
"active_params_b": {
"type": "number",
"exclusiveMinimum": 0,
"description": "Parameters read per token in billions. Equals total for dense models. Sets speed, never memory."
},
"is_moe": {
"type": "boolean"
}
},
"required": [
"id",
"name",
"total_params_b",
"active_params_b",
"is_moe"
],
"additionalProperties": false
},
"hardware": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$"
},
"name": {
"type": "string",
"maxLength": 120
},
"memory_gb": {
"type": "number",
"exclusiveMinimum": 0
},
"memory_kind": {
"type": "string",
"enum": [
"dedicated",
"unified",
"shared-system"
]
},
"bandwidth_gb_s": {
"type": "number",
"exclusiveMinimum": 0,
"description": "Theoretical peak memory bandwidth."
},
"status": {
"type": "string",
"enum": [
"current",
"eol",
"used-market",
"announced"
],
"description": "announced means not yet on sale; eol means no longer made."
}
},
"required": [
"id",
"name",
"memory_gb",
"memory_kind",
"bandwidth_gb_s",
"status"
],
"additionalProperties": false
},
"memory": {
"type": "object",
"properties": {
"weights_gb": {
"type": "number",
"minimum": 0,
"description": "All parameters quantised. For a Mixture-of-Experts model this counts every expert."
},
"kv_cache_gb": {
"type": "number",
"minimum": 0
},
"overhead_gb": {
"type": "number",
"minimum": 0,
"description": "Fixed framework overhead."
},
"needed_gb": {
"type": "number",
"minimum": 0,
"description": "weights + KV cache + overhead."
},
"available_gb": {
"type": "number",
"minimum": 0,
"description": "Memory usable for the model after any operating-system reserve."
},
"headroom_gb": {
"type": "number",
"description": "available - needed. Negative when the model does not fit."
},
"offloaded_gb": {
"type": "number",
"minimum": 0,
"description": "Weights that would live in system RAM. 0 when fully resident."
},
"confidence": {
"type": "string",
"enum": [
"measured",
"estimated",
"community"
]
},
"kv_basis": {
"type": "string",
"enum": [
"published_architecture",
"inferred_architecture"
],
"description": "Whether the KV cache is computed from a published config or inferred from the parameter count."
}
},
"required": [
"weights_gb",
"kv_cache_gb",
"overhead_gb",
"needed_gb",
"available_gb",
"headroom_gb",
"offloaded_gb",
"confidence",
"kv_basis"
],
"additionalProperties": false,
"description": "Memory arithmetic for the requested quantisation and context length."
},
"speed": {
"anyOf": [
{
"type": "object",
"properties": {
"low_tok_s": {
"type": "number",
"minimum": 0
},
"high_tok_s": {
"type": "number",
"minimum": 0
},
"confidence": {
"type": "string",
"enum": [
"measured",
"estimated",
"community"
]
},
"basis": {
"anyOf": [
{
"type": "string",
"maxLength": 200
},
{
"type": "null"
}
],
"description": "Where the figure comes from, e.g. a benchmark label. null for a modelled estimate."
},
"sample_count": {
"anyOf": [
{
"type": "integer",
"exclusiveMinimum": 0,
"maximum": 9007199254740991
},
{
"type": "null"
}
],
"description": "Number of measured runs behind a measured or community figure."
},
"last_verified": {
"anyOf": [
{
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$"
},
{
"type": "null"
}
]
}
},
"required": [
"low_tok_s",
"high_tok_s",
"confidence",
"basis",
"sample_count",
"last_verified"
],
"additionalProperties": false,
"description": "Decode speed in tokens per second, always a range with a confidence label."
},
{
"type": "null"
}
],
"description": "null when the model does not fit, because a speed for hardware that cannot load the weights would be invented."
},
"notes": {
"maxItems": 8,
"type": "array",
"items": {
"type": "string",
"maxLength": 300
}
},
"sources": {
"maxItems": 8,
"type": "array",
"items": {
"type": "string",
"pattern": "^https:\\/\\/"
},
"description": "Primary sources for the model and hardware records. May be empty."
},
"url": {
"type": "string",
"pattern": "^https:\\/\\/llmconfigurator\\.com\\/en\\/"
}
},
"required": [
"data_as_of",
"summary",
"verdict",
"quantisation",
"context_length",
"model",
"hardware",
"memory",
"speed",
"notes",
"sources",
"url"
],
"additionalProperties": false
}🟢models_for_hardware(hardware_id, quantisation, context_length, use_case, limit)
Use this when the user asks what models a given machine can run. Needs a hardware id from search_catalog. Returns up to 20 catalogue models ranked for that machine, each with a fit verdict and a speed range, optionally filtered to one use such as coding. It covers only models in the catalogue, at one quantisation and context length per call, and leaves out models that do not fit.
输入模式
{
"type": "object",
"properties": {
"hardware_id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$",
"description": "Hardware id from search_catalog."
},
"quantisation": {
"default": "Q4_K_M",
"description": "Weight quantisation. One of Q2_K, Q3_K_M, Q4_K_M, Q5_K_M, Q6_K, Q8_0, F16. Default Q4_K_M.",
"type": "string",
"enum": [
"Q2_K",
"Q3_K_M",
"Q4_K_M",
"Q5_K_M",
"Q6_K",
"Q8_0",
"F16"
]
},
"context_length": {
"default": 4096,
"description": "Context window in tokens that the KV cache is sized for. 512 to 262144. Default 4096.",
"type": "integer",
"minimum": 512,
"maximum": 262144
},
"use_case": {
"description": "Only list models tagged for this use. One of coding, chat, reasoning, agents, vision. Omit for all models.",
"type": "string",
"enum": [
"coding",
"chat",
"reasoning",
"agents",
"vision"
]
},
"limit": {
"default": 10,
"description": "Maximum models, 1 to 20. Default 10.",
"type": "integer",
"minimum": 1,
"maximum": 20
}
},
"required": [
"hardware_id"
]
}输出模式
{
"type": "object",
"properties": {
"data_as_of": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$",
"description": "Date the bundled catalogue was last updated. Not the time of this call."
},
"summary": {
"type": "string",
"maxLength": 600,
"description": "One or two plain-language sentences stating the answer."
},
"quantisation": {
"type": "string",
"enum": [
"Q2_K",
"Q3_K_M",
"Q4_K_M",
"Q5_K_M",
"Q6_K",
"Q8_0",
"F16"
]
},
"context_length": {
"type": "integer",
"minimum": -9007199254740991,
"maximum": 9007199254740991
},
"use_case": {
"anyOf": [
{
"type": "string",
"enum": [
"coding",
"chat",
"reasoning",
"agents",
"vision"
]
},
{
"type": "null"
}
],
"description": "The use-case filter applied, or null."
},
"hardware": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$"
},
"name": {
"type": "string",
"maxLength": 120
},
"memory_gb": {
"type": "number",
"exclusiveMinimum": 0
},
"memory_kind": {
"type": "string",
"enum": [
"dedicated",
"unified",
"shared-system"
]
},
"bandwidth_gb_s": {
"type": "number",
"exclusiveMinimum": 0,
"description": "Theoretical peak memory bandwidth."
},
"status": {
"type": "string",
"enum": [
"current",
"eol",
"used-market",
"announced"
],
"description": "announced means not yet on sale; eol means no longer made."
}
},
"required": [
"id",
"name",
"memory_gb",
"memory_kind",
"bandwidth_gb_s",
"status"
],
"additionalProperties": false
},
"models": {
"type": "array",
"items": {
"type": "object",
"properties": {
"rank": {
"type": "integer",
"exclusiveMinimum": 0,
"maximum": 9007199254740991
},
"model_id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$"
},
"name": {
"type": "string",
"maxLength": 120
},
"verdict": {
"type": "string",
"enum": [
"fits_comfortably",
"fits_tight",
"needs_offload",
"does_not_fit"
]
},
"needed_gb": {
"type": "number",
"minimum": 0
},
"speed": {
"anyOf": [
{
"type": "object",
"properties": {
"low_tok_s": {
"type": "number",
"minimum": 0
},
"high_tok_s": {
"type": "number",
"minimum": 0
},
"confidence": {
"type": "string",
"enum": [
"measured",
"estimated",
"community"
]
},
"basis": {
"anyOf": [
{
"type": "string",
"maxLength": 200
},
{
"type": "null"
}
],
"description": "Where the figure comes from, e.g. a benchmark label. null for a modelled estimate."
},
"sample_count": {
"anyOf": [
{
"type": "integer",
"exclusiveMinimum": 0,
"maximum": 9007199254740991
},
{
"type": "null"
}
],
"description": "Number of measured runs behind a measured or community figure."
},
"last_verified": {
"anyOf": [
{
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$"
},
{
"type": "null"
}
]
}
},
"required": [
"low_tok_s",
"high_tok_s",
"confidence",
"basis",
"sample_count",
"last_verified"
],
"additionalProperties": false,
"description": "Decode speed in tokens per second, always a range with a confidence label."
},
{
"type": "null"
}
]
},
"url": {
"type": "string",
"pattern": "^https:\\/\\/llmconfigurator\\.com\\/en\\/"
}
},
"required": [
"rank",
"model_id",
"name",
"verdict",
"needed_gb",
"speed",
"url"
],
"additionalProperties": false
}
},
"url": {
"type": "string",
"pattern": "^https:\\/\\/llmconfigurator\\.com\\/en\\/"
}
},
"required": [
"data_as_of",
"summary",
"quantisation",
"context_length",
"use_case",
"hardware",
"models",
"url"
],
"additionalProperties": false
}🟢compare_hardware(hardware_ids)
Use this when the user compares two to four machines. Needs hardware ids from search_catalog. Returns memory, bandwidth, memory type, release year, status and a dated price where one is recorded, side by side. A price of null means none is on record. It does not check any model against the machines; use check_hardware_fit for that.
输入模式
{
"type": "object",
"properties": {
"hardware_ids": {
"minItems": 2,
"maxItems": 4,
"type": "array",
"items": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$"
},
"description": "Two to four hardware ids from search_catalog."
}
},
"required": [
"hardware_ids"
]
}输出模式
{
"type": "object",
"properties": {
"data_as_of": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$",
"description": "Date the bundled catalogue was last updated. Not the time of this call."
},
"summary": {
"type": "string",
"maxLength": 600,
"description": "One or two plain-language sentences stating the answer."
},
"hardware": {
"minItems": 2,
"maxItems": 4,
"type": "array",
"items": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$"
},
"name": {
"type": "string",
"maxLength": 120
},
"memory_gb": {
"type": "number",
"exclusiveMinimum": 0
},
"memory_kind": {
"type": "string",
"enum": [
"dedicated",
"unified",
"shared-system"
]
},
"bandwidth_gb_s": {
"type": "number",
"exclusiveMinimum": 0,
"description": "Theoretical peak memory bandwidth."
},
"status": {
"type": "string",
"enum": [
"current",
"eol",
"used-market",
"announced"
],
"description": "announced means not yet on sale; eol means no longer made."
},
"class": {
"type": "string",
"maxLength": 40
},
"architecture": {
"type": "string",
"maxLength": 80
},
"release_year": {
"anyOf": [
{
"type": "integer",
"minimum": -9007199254740991,
"maximum": 9007199254740991
},
{
"type": "null"
}
]
},
"system_ram_gb": {
"type": "number",
"minimum": 0,
"description": "System RAM available for offload. 0 for unified-memory machines."
},
"price": {
"anyOf": [
{
"type": "object",
"properties": {
"usd": {
"type": "number",
"exclusiveMinimum": 0
},
"as_of": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$",
"description": "Date this price was checked."
},
"market": {
"type": "string",
"enum": [
"new",
"used",
"refurb"
]
},
"stale": {
"type": "boolean",
"description": "True when the price is older than 90 days."
},
"source": {
"type": "string",
"maxLength": 200
}
},
"required": [
"usd",
"as_of",
"market",
"stale",
"source"
],
"additionalProperties": false
},
{
"type": "null"
}
],
"description": "null when this catalogue holds no dated price. Never 0."
},
"notes": {
"anyOf": [
{
"type": "string",
"maxLength": 400
},
{
"type": "null"
}
]
},
"sources": {
"maxItems": 8,
"type": "array",
"items": {
"type": "string",
"pattern": "^https:\\/\\/"
}
},
"url": {
"type": "string",
"pattern": "^https:\\/\\/llmconfigurator\\.com\\/en\\/"
}
},
"required": [
"id",
"name",
"memory_gb",
"memory_kind",
"bandwidth_gb_s",
"status",
"class",
"architecture",
"release_year",
"system_ram_gb",
"price",
"notes",
"sources",
"url"
],
"additionalProperties": false
}
}
},
"required": [
"data_as_of",
"summary",
"hardware"
],
"additionalProperties": false
}🟢get_model_specs(model_id)
Use this when the user asks about one model's size, architecture, context window or licence. Needs a model id from search_catalog. Returns total and active parameters, weight size per quantisation, context length, licence, release date and sources. It does not say whether the model fits any machine; use check_hardware_fit for that.
输入模式
{
"type": "object",
"properties": {
"model_id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$",
"description": "Model id from search_catalog."
}
},
"required": [
"model_id"
]
}输出模式
{
"type": "object",
"properties": {
"data_as_of": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$",
"description": "Date the bundled catalogue was last updated. Not the time of this call."
},
"summary": {
"type": "string",
"maxLength": 600,
"description": "One or two plain-language sentences stating the answer."
},
"model": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80,
"pattern": "^[a-z0-9][a-z0-9._-]*$"
},
"name": {
"type": "string",
"maxLength": 120
},
"total_params_b": {
"type": "number",
"exclusiveMinimum": 0,
"description": "Total parameters in billions. Sets memory residency."
},
"active_params_b": {
"type": "number",
"exclusiveMinimum": 0,
"description": "Parameters read per token in billions. Equals total for dense models. Sets speed, never memory."
},
"is_moe": {
"type": "boolean"
},
"family": {
"type": "string",
"maxLength": 120
},
"context_length": {
"anyOf": [
{
"type": "integer",
"exclusiveMinimum": 0,
"maximum": 9007199254740991
},
{
"type": "null"
}
],
"description": "The model's default context window in tokens."
},
"license": {
"anyOf": [
{
"type": "string",
"maxLength": 80
},
{
"type": "null"
}
]
},
"release_date": {
"anyOf": [
{
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}$"
},
{
"type": "null"
}
]
},
"status": {
"type": "string",
"enum": [
"current",
"legacy",
"deprecated"
]
}
},
"required": [
"id",
"name",
"total_params_b",
"active_params_b",
"is_moe",
"family",
"context_length",
"license",
"release_date",
"status"
],
"additionalProperties": false
},
"quantisations": {
"type": "array",
"items": {
"type": "object",
"properties": {
"quantisation": {
"type": "string",
"enum": [
"Q2_K",
"Q3_K_M",
"Q4_K_M",
"Q5_K_M",
"Q6_K",
"Q8_0",
"F16"
]
},
"bits_per_weight": {
"type": "number",
"exclusiveMinimum": 0
},
"weights_gb": {
"type": "number",
"minimum": 0
},
"confidence": {
"type": "string",
"enum": [
"measured",
"estimated",
"community"
]
}
},
"required": [
"quantisation",
"bits_per_weight",
"weights_gb",
"confidence"
],
"additionalProperties": false
},
"description": "Weight size per quantisation. Excludes KV cache and overhead, which depend on context length."
},
"sources": {
"maxItems": 8,
"type": "array",
"items": {
"type": "string",
"pattern": "^https:\\/\\/"
}
},
"url": {
"type": "string",
"pattern": "^https:\\/\\/llmconfigurator\\.com\\/en\\/"
}
},
"required": [
"data_as_of",
"summary",
"model",
"quantisations",
"sources",
"url"
],
"additionalProperties": false
}社区
证据
最近观测
已验证未记录版本5 个工具