Loading ToolBench
MCP server for VTEX IO development — Store Framework, React, Node services, GraphQL, Admin and more
Server has 9 tools with complete schemas and descriptions. Naming follows verb_noun convention (lookup-, add-, scaffold-, search-, explain-). Descriptions are substantive (100-250 chars), exceeding the 10-char minimum. However, parameter descriptions lack actionable constraints (no enums for builder types, no format specs for appName/vendor). Output schemas are undocumented, LLMs cannot predict response structure. Error handling is absent from visible code. No tool annotations (readOnlyHint, idempotentHint). Composition is sound: tools are single-purpose and chainable (search-concepts → explain-concept). Security: no secrets in params (good). Overall: solid foundation, but missing LLM-optimization details that would push this to 75+.
Generate a JSONC snippet for a VTEX IO block ready to paste into blocks.jsonc. Validates props against the block schema.
Get full documentation for a VTEX concept by exact ID (e.g. "vtex-io-builders", "vtex-io-css-handles", "vtex-io-service-json"). Use search-concepts to find the right ID first.
Look up the props, description, and usage examples for any VTEX IO Store Framework block (e.g. rich-text, flex-layout.row, shelf, image, info-card)
Look up VTEX REST API reference (endpoints, auth, request/response models) for any VTEX API: catalog, orders, checkout, master-data, logistics, pricing, intelligent-search, session, headless-cms, promotions, payments-gateway, license-manager
Generate a GraphQL schema and resolver files for a VTEX IO app, including schema.graphql and TypeScript resolver stubs for queries and mutations
Generate a complete Node.js service scaffold for VTEX IO including index.ts, service.json, clients, and middleware handlers for specified routes and events
Output schemas undocumented. LLMs cannot predict response structure (fields, types, pagination). Blocks downstream tool chaining and forces agents to guess at result format.
Parameter constraints missing. 'builders' enum is declared in schema but not in description; 'appName' and 'vendor' lack format specs (kebab-case, length, allowed chars). LLMs cannot self-validate input.
No error handling guidance. Tools return success/failure but do not guide LLM on recovery (retry, lookup, ask user). Agents cannot self-correct on invalid input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Generate a complete VTEX IO app scaffold with manifest.json and all folder structures for the selected builders (store, react, node, graphql, styles, messages, admin, pixel)
Search across 391 VTEX documentation files by keywords. Returns ranked results with excerpts. Use this to find concepts before calling explain-concept.
Search for a term across all VTEX IO official courses (basic-blocks, layout-blocks, styles-course, store-block, service-course, admin, calling-commerce-apis, content-workflow, store-performance). Returns excerpts with surrounding context.
No tool annotations. Tools lack readOnlyHint (all are read-only), idempotentHint (scaffolding is idempotent), or destructiveHint. Agents cannot reason about side effects or retry safety.
Parameter descriptions lack dependency hints. 'explain-concept' says 'Use search-concepts to find the right ID first' but search-concepts does not document that it returns concept IDs. Breaks tool chaining discovery.