Measured local-LLM tok/s on home AI rigs. Search builds, can-it-run, best GPU; sourced + linked.
Scored across 4 tools
Each tool answers a distinct question: search builds, fetch a build by slug, recommend the fastest GPU for a model, and check compatibility for a given setup. The only mild overlap is between best_gpu_for_model and search_builds when querying measured results for a model, but the phrasing and outputs differ enough that selection should be clear.
Two tools follow a verb_noun pattern (get_build, search_builds), while best_gpu_for_model is noun-phrase and can_it_run is a question-style name. The mix is readable and fairly intuitive, but it is not a single predictable convention.
Four tools is well-scoped for a narrow benchmark lookup service: search, detail, recommendation, and compatibility check. Each tool has a clear role and no obvious filler.
The surface covers the main read-only workflows for measured builds: discovery, detail, best-for-model, and can-it-run. Minor gaps include no explicit way to enumerate supported models/GPUs or compare multiple builds side by side, but agents can mostly work around this through search and get_build.
Return the fastest measured RigBase result for a model, plus tok/s per euro where a build cost or known GPU street price exists. Includes source credit and canonical URL.
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No | How many top results to return (default 5, max 15). | |
| model | Yes | Model name or substring, e.g. "Gemma 4 31B" or "Qwen3.8". |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and openWorldHint=false, so safety is covered. The description goes beyond them by disclosing a conditional behavior — the cost-efficiency figure only appears 'where a build cost or known GPU street price exists' — and by noting the response carries source credit and a canonical URL.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, zero waste, with the primary output (fastest measured result) front-loaded and the conditional metric following. Nothing to trim.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description usefully sketches the return payload (fastest result, tok/s per euro, source credit, canonical URL). A caller still doesn't know the result object's shape, but for a simple two-parameter read tool this is close to sufficient.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, with both `model` and `limit` fully documented in the schema, so the schema does the heavy lifting. The description adds no syntax or format detail (e.g., substring matching behavior) beyond what is already structured.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource ('Return the fastest measured RigBase result for a model') and adds the derived metric scope (tok/s per euro). It is clear what the tool produces, though it never names the sibling tools it differs from.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies the use case (finding the best GPU for a given model) but gives no explicit when-to-use/when-not guidance or routing to can_it_run, get_build, or search_builds. Selection between this and its siblings is left to inference.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Ask whether a given GPU setup can run a model. Returns measured RigBase results when available; otherwise a VRAM-fit estimate clearly labelled as an estimate (weights arithmetic, not a measured speed). Advises against quants below Q4. Always includes source credit and URLs for measured rows.
| Name | Required | Description | Default |
|---|---|---|---|
| gpu | Yes | GPU or system name, e.g. "RTX 3090", "RTX 5090", "DGX Spark", "Apple Silicon Mac". | |
| count | No | How many identical GPUs / nodes (default 1). | |
| model | Yes | Model name, e.g. "Qwen3.8 27B", "gpt-oss 120B", "Gemma 4 31B". | |
| quant | No | Optional quant for the estimate (default Q4_K_M). Avoid below Q4. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already cover the safety profile (readOnlyHint, closed-world), and the description adds genuinely non-obvious behavior: measured RigBase results when available, otherwise a labelled VRAM-fit estimate, plus source credit and URLs for measured rows. This tells the agent how much to trust the answer, which annotations and schema cannot. It stops short of saying what happens when the GPU or model is unknown.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Three tight sentences, front-loaded with the purpose, then the measured-vs-estimate distinction, then the quant advice. No filler and nothing padded.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description does the necessary work of explaining what comes back: measured rows with attribution or a clearly labelled estimate. That is sufficient for an agent to call it correctly, though error/unknown-input behavior and any rate or lookup limits are unaddressed.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so all four parameters are already documented with examples and defaults. The description only reinforces the quant guidance ("advises against quants below Q4"), which the schema's quant field already states. Baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb and resource ("Ask whether a given GPU setup can run a model"), which is clearly the inverse direction of the sibling best_gpu_for_model. However, it never names that sibling or otherwise explicitly distinguishes itself, so the differentiation is inferable rather than stated.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The framing "Ask whether..." implies a compatibility-check use case, but there is no explicit when-to-use or when-not-to-use guidance and no mention of best_gpu_for_model / search_builds as alternatives. An agent must infer that this is the GPU-then-model direction rather than the model-then-GPU direction.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Fetch one RigBase build by slug (e.g. "dual-3090-ryzen-3700x"). Returns parts, measured results, source credit and canonical URL.
| Name | Required | Description | Default |
|---|---|---|---|
| slug | Yes | Build slug from a RigBase /builds/.../ URL. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and openWorldHint=false, so safety is covered and the description doesn't contradict them. It adds real behavioral value by disclosing the return contents and the fact that this is a single-record lookup keyed by slug. It stops short of error behavior (unknown slug) or whether results are cached/live.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, front-loaded with the verb and cardinality, followed by the return summary. Every clause carries information an agent needs; nothing is restated from the name or title.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a one-parameter read tool with no output schema, the description covers the essentials: what it fetches, the key format, and what the response contains. The only gap is the absence of any note on failure when a slug doesn't exist, which matters for a slug-keyed lookup.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the baseline is 3, but the description goes beyond it with a concrete literal example ("dual-3090-ryzen-3700x") that pins down the expected slug format. Combined with the schema's note that the slug is derived from a /builds/.../ URL, the agent has both provenance and shape.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
"Fetch one RigBase build by slug" names a specific verb (fetch), resource (RigBase build), and cardinality (one), which cleanly separates it from the sibling search_builds. The second sentence enumerates the payload (parts, measured results, source credit, canonical URL), so the agent knows exactly what it gets back.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Usage is only implied: the description says a slug is required and the schema says the slug comes from a /builds/.../ URL, which tells the agent the precondition but not the routing rule. It never names search_builds or best_gpu_for_model as the alternative for discovery workflows, so an agent must infer that this is the follow-up call after a search.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search RigBase measured home AI builds. Filter by GPU name, model name, minimum VRAM (GB), and/or maximum build cost. Returns top matches with tok/s, quant, engine, source credit and canonical getrigbase.com URL. Never invents numbers.
| Name | Required | Description | Default |
|---|---|---|---|
| gpu | No | Substring to match in the GPUs field (e.g. "3090", "DGX Spark", "5090"). | |
| limit | No | Max builds to return (default 8, max 25). | |
| model | No | Substring to match a measured model name (e.g. "Qwen3.8", "gpt-oss 120B"). | |
| currency | No | Optional currency filter: EUR or USD. | |
| max_price | No | Maximum reported build cost (same currency as the build, usually EUR or USD). | |
| min_vram_gb | No | Minimum total VRAM / unified memory in GB. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already establish readOnlyHint=true and openWorldHint=false, so safety is covered. The description goes further by naming the return payload (tok/s, quant, engine, source credit, canonical URL) and asserting 'Never invents numbers', which is a real reliability trait an agent can reason about. It stops short of describing ordering or truncation behavior.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Three short sentences with the purpose and filter set front-loaded, then the return shape. Minimal waste; only the closing 'Never invents numbers' is slightly promotional, though it still carries a trust signal.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description does the necessary work of listing returned fields, and it covers the filter dimensions. Gaps are minor: no explanation of result ranking ('top matches') or of how unpriced builds affect max_price filtering.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so all six parameters are already documented with examples and defaults. The description restates four of the six filters (gpu, model, min_vram_gb, max_price) but adds no syntax or semantics beyond the schema, and omits limit and currency entirely. Baseline 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource ('Search RigBase measured home AI builds') and enumerates the filterable dimensions, so the agent knows exactly what corpus is being queried. It does not, however, differentiate itself from siblings like get_build or best_gpu_for_model.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The filter list implies the use case (broad lookup over measured builds), but there is no explicit statement of when to choose this over get_build for a known build or best_gpu_for_model for a recommendation. Usage is inferable rather than stated.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
best_gpu_for_modelcan_it_runget_buildsearch_buildsRead-only: which local LLMs a GPU or Mac can run - VRAM fit, tokens/sec, model specs.
Can this LLM run on my GPU? VRAM, speed ceiling and what fits instead, for any model and GPU.
GPU and LLM inference benchmarks, hardware evidence, deployment recommendations, and launch configs.
Will a local LLM run on your hardware? GGUF quant, buy-vs-rent-vs-API cost, used-GPU prices.
Add one secure layer between your agents and this server.
