Loading ToolBench

Reading the current corpus…

frete-mcp MCP server · F · ToolBench

frete-mcp

MCP server for frete.center — Brazilian freight (TURBO 3h, drivers, tracking) as a native tool in Claude Code, Claude Desktop and Cursor.

2026-07-28+STDIO6 tools

49/100

Definition Quality55

Protocol Readiness50

Supportability39

Spec posture

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

Static source inference · medium confidence

Current-spec patterns detected

Summary

Server defines 6 tools with Portuguese names and descriptions. All tools have input schemas with type definitions and descriptions. However, critical gaps exist: (1) Tool names do not follow verb_noun convention (e.g., 'frete_cotar' should be 'get_quote' or 'quote_freight'); (2) Descriptions are adequate (100-200 chars) but lack actionable context for LLM selection; (3) Output schemas are NOT documented, callers cannot predict response structure; (4) Error handling is minimal, HTTP errors are caught but recovery guidance is absent; (5) Parameters lack validation constraints (e.g., CEP format, km range). The frete_tracking tool is particularly weak: it returns a public URL instead of structured tracking data, forcing the LLM to reason about HTML content. Overall, definitions are present but lack production-grade polish.

Tools (6)

frete_carteiraread onlyauthsource verified55/100

Saldo + extrato da carteira do tenant autenticado. Requer FRETE_BEARER.

frete_cotarread onlysource verified60/100

Cota frete urbano via Lalamove entre dois CEPs brasileiros. Retorna preço, distância, ETA e tier (TURBO 3h ou EXPRESS 12h). Não exige autenticação.

frete_meus_pedidosread onlyauthsource verified55/100

Lista pedidos do tenant autenticado. Requer FRETE_BEARER no env. Para gerar token: POST /api/app/v1/auth/login.

frete_motoristas_proximosread onlysource verified55/100

Lista motoristas online (com toggle "disponível agora") em raio de N km de um CEP ou ponto. Coordenadas com jitter ~300m por privacidade LGPD. Cache 30s.

frete_snapshotread onlysource verified55/100

Snapshot agregado da plataforma frete.center: contagem de motoristas online, CEPs cacheados, distâncias proprietárias, blog posts, rotas SEO, marketplace open.

frete_trackingread onlysource verified47/100

Busca status atual + timeline de um pedido pelo código de tracking público (8 chars maiúsculos).

Issues (5)

HIGHtool ↗

Tool names do not follow verb_noun convention. Names like 'frete_cotar', 'frete_motoristas_proximos' are domain-specific but lack action verbs (get_, list_, search_). LLMs infer intent from verb prefixes; these names force the LLM to read descriptions to understand what happens when called.

frete_cotarfrete_motoristas_proximosfrete_trackingfrete_snapshotfrete_meus_pedidosfrete_carteira

HIGHtool ↗

Output schemas are not documented. The code calls http() and returns raw API responses without documenting what fields the LLM should expect. For example, frete_cotar returns price/distance/ETA but the schema is not declared. LLMs cannot plan downstream calls or extract data reliably.

frete_cotarfrete_motoristas_proximosfrete_snapshotfrete_meus_pedidosfrete_carteira

HIGHresponse-shaper ↗

frete_tracking returns a public URL instead of structured tracking data. The tool description promises 'status atual + timeline' but the implementation returns only a URL and a note. LLMs cannot extract actionable tracking info without parsing HTML.

Recommendations

Rename tools to follow verb_noun convention: 'frete_cotar' → 'get_freight_quote', 'frete_motoristas_proximos' → 'list_nearby_drivers', 'frete_tracking' → 'get_tracking_status', 'frete_snapshot' → 'get_platform_snapshot', 'frete_meus_pedidos' → 'list_my_orders', 'frete_carteira' → 'get_wallet_balance'. This makes intent obvious from the name alone.
Document output schemas for all tools. Add a 'returns' field to each Tool interface describing the response structure (e.g., { price: number, distance_km: number, eta_minutes: number, tier: string }). Include this in tool descriptions or as a separate schema field.
Fix frete_tracking to return structured data. Instead of a public URL, call an internal API endpoint (or parse the HTML server-side) and return { code, status, current_location, eta, timeline: [{timestamp, event, location}] }. If no JSON endpoint exists, add one to the backend.
Add JSON Schema constraints to parameters: CEP fields should have minLength/maxLength 8 and pattern '^[0-9]{8}$'; km should have minimum 1, maximum 50; service_type and tier are already enums (good). Update inputSchema objects in the Tool definitions.
Enhance error handling with recovery guidance. Catch specific error conditions (404 not found, 401 unauthorized, 400 bad request) and return actionable messages: 'CEP not found. Verify the 8-digit format (e.g., 01310100).' or 'FRETE_BEARER not configured. Set the environment variable and restart.' Use error classification (retryable, user-fixable, fatal) to guide LLM behavior.

Score history

Overall score trend

First recorded score · v2 rubric

49/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	49	2026-07-28+	v2