Loading ToolBench

Reading the current corpus…

token-scout MCP server · F · ToolBench

token-scout

MCP server that finds free LLM models across providers. Single MCP tool. Zero inference overhead — returns endpoints, you call models directly. No proxy, no middleman.

2026-07-28+STDIO6 tools

41/100

Definition Quality42

Protocol Readiness48

Supportability36

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Token Scout has 6 tools with mixed quality. Tool names lack action verbs (scout, status, consume, reset, budget, discover are acceptable but not optimal verb-noun patterns). Descriptions exist but are verbose and lack LLM optimization, scout's description is 289 chars, exceeding the 194-char production baseline and burying key details. Input schemas are partially visible in the code (scout has detailed schema, consume/reset have minimal params), but several tools show incomplete parameter documentation. The codebase shows Python and Rust implementations, Rust edition is listed as '2024' (invalid; should be 2021 or earlier), suggesting incomplete/exploratory code. Output schemas are not documented. Error handling is absent from visible code. The server is STDIO-only, which is a hard cap at protocol readiness score 50.

Tools (6)

budgetread only32/100

Get budget advice based on Claude usage data from /tmp/claude-usage.json. Analyzes session and weekly budget consumption to recommend 'burn_claude', 'conserve', or 'free_only' strategy.

consumewritesource verified55/100

Record quota consumption for a provider/model pair. Updates runtime in-memory quota tracker (resets daily).

discoverread onlysource verified38/100

Discover available models from Ollama constellation (local machines) and OpenRouter live (cloud providers). Populates the registry with dynamically discovered models.

resetwrite37/100

Reset all quota tracking data. Clears the in-memory quota tracker.

scoutread onlysource verified57/100

Search for LLM models across configured providers by query and ranking preference. Supports hard constraints via require (reasoning_format, tool_format, tool_reliability, min_context, min_completion, modality). Respects TOKEN_SCOUT_MAX_COST environment variable for cost filtering.

statusread only33/100

Get status of all configured providers and available models. Equivalent to scout with empty query.

Issues (6)

CRITICALtool ↗

Multiple tools lack input schemas or have incomplete parameter definitions (status, reset, budget, discover have no visible schemas; consume has minimal param documentation)

statusresetbudgetdiscover

HIGHtool-description ↗

Tool descriptions are verbose and lack LLM optimization. Scout description is 289 chars (exceeds 194-char baseline); status is vague ('Equivalent to scout with empty query' forces agent to read scout docs); budget relies on undocumented file path /tmp/claude-usage.json

scoutstatusbudget

HIGHresponse-shaper ↗

No documented output schemas. Agents cannot determine what fields to expect from scout, status, budget, or discover results. This blocks multi-step planning and forces agents to parse unstructured output

scoutstatusbudgetdiscover

Recommendations

Add explicit JSON Schema output declarations for all 6 tools. Example for scout: {"type": "array", "items": {"type": "object", "properties": {"provider": {"type": "string"}, "model": {"type": "string"}, "model_name": {"type": "string"}, ...}}}
Rewrite tool descriptions to 50-150 chars, LLM-optimized. Scout: 'Find LLM models by query and constraints (context, reasoning format, tool support).' Status: 'List all models and providers with current status.' Budget: 'Get spending advice based on Claude usage logs.' Remove verbose constraint lists from descriptions; they belong in parameter constraints.
Add input schemas for status, reset, budget, discover. Even empty-param tools should declare {} with a note. For discover: document that it populates registry from Ollama/OpenRouter discovery endpoints, what does it return, what providers does it hit?
Replace empty string '' in scout's prefer enum with null or a named value like 'relevance'. Use proper JSON Schema. Clarify scout's 'require' parameter structure with explicit nested schema.
Document output schemas for all list/search tools. If scout returns paginated results, include total_count, next_cursor, and per-model fields (provider, model, context_tokens, cost_per_1k, etc.). If status returns provider + model arrays, show that structure.
Add error handling guidance to descriptions. Example for scout: 'Returns empty array if no models match constraints. If all providers lack API keys, returns [] with note in separate field.' For consume: 'Returns error if provider/model unknown; call status first to validate.'

Score history

Overall score trend

First recorded score · v2 rubric

41/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	41	2026-07-28+	v2