Open spec, scorer, and MCP generator for the agent-driven web. Monorepo containing multiple MCP servers: Commerce (agent-driven commerce), Middleware (proxy for any website), SEO (agent discovery analytics), and tools for scoring agent readiness and generating MCP servers from API specs.
AgentWeb provides 16 tools with reasonable descriptions and parameter schemas. Most tools have clear names (verb_noun pattern) and documented inputs. However, output schemas are not visible in the provided source code, parameter descriptions lack depth (missing constraints, ranges, format specs), and error handling guidance is absent. Tool descriptions average ~120 chars (below the 194-char baseline for A+ tools). No evidence of idempotency hints, confirmation patterns for irreversible ops, or recovery guidance in error responses. Commerce tools (checkout, negotiate) are high-risk but lack dry-run or confirmation mechanisms.
Add a product to a cart with specified quantity and attributes (size, color, etc.).
Apply a promotional code to a cart to receive a discount.
Browse and read content from the site. Returns the main text content of a page. Use this to explore the site, read articles, view product pages, or find information.
Complete a purchase from a cart. Requires shipping address and payment information.
Compare your agent readiness against competitors. Shows where you lead and where you're behind.
Create a new shopping cart for the agent to add items to.
Extract structured product data from the site. Returns products with pricing, availability, and attributes.
Output schemas not documented. Tool descriptions state what is returned (e.g., 'Returns product details including availability, pricing, and ratings') but the actual response structure is not visible in source code. LLMs cannot plan downstream tool calls or extract specific fields without knowing the response schema.
Irreversible operations (checkout, negotiate) lack confirmation or dry-run patterns. Agents can trigger purchases or price negotiations without a confirmation step, risking unintended side effects. No evidence of idempotentHint or destructiveHint annotations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2025-06-18+ | v2 |
Get a prioritized list of actions to improve agent visibility. Each recommendation includes impact assessment and implementation effort.
Get brand policies including return policy, shipping policy, warranty, and other business rules.
Get detailed information about a specific product by ID, including full description, pricing, availability, ratings, and related products.
See what agents are asking about your brand — top queries, mention rate, sentiment, and platform breakdown.
Get the agent visibility score for the site. Shows how well AI agents can discover and recommend your brand across 5 dimensions: discoverability, data quality, policy clarity, brand consistency, and competitive position.
Negotiate pricing or terms for a product or order. Agents can propose alternative prices, bundles, or payment terms.
Remove a product from a cart.
Search the product catalog. Supports natural language queries, category filtering, price ranges, size/color attributes, and occasion-based search. Returns product details including availability, pricing, and ratings.
Search the site's indexed content by keyword. Returns matching pages with titles, descriptions, and relevance scores.
Parameter descriptions lack actionable constraints. E.g., 'limit' in search_products says 'Max results (default: 5, max: 20)' but no enum, pattern, or range validation is visible. 'attributes' in add_to_cart is described as 'Selected attributes (e.g., {size: "M", color: "Navy"})', example values in descriptions invite literal reuse by LLMs.
No error handling guidance visible. Tool descriptions do not explain what to do if a product is out of stock, a promo code is invalid, or checkout fails. LLMs receive no recovery hints (e.g., 'Try search_products() with different filters' or 'Invalid code, call get_policies() to see active promotions').
Tool composition gaps. 'create_cart' returns no output schema, so the agent cannot know the cart_id to pass to add_to_cart. 'search_products' does not document whether it returns product_id (needed for get_product and add_to_cart). Broken chains force discovery detours.