Loading ToolBench
Model Context Protocol (MCP) server for Yandex Webmaster — query indexing, search queries, diagnostics and backlinks, and submit URLs for recrawl through an AI agent.
Server has 8 tools with mostly complete schemas and descriptions. Naming follows verb_noun convention (get_*, list_*, submit_*). Descriptions are detailed and contextual (avg ~180 chars), exceeding baseline. All visible tools have input schemas with type definitions and parameter descriptions. However, 2 tools (get_search_queries, list_sitemaps) lack visible source code, only file paths provided, preventing full schema verification. Tool descriptions could better explain when to call each vs. alternatives. Error handling is present but generic. Output schemas are not explicitly documented in the provided code.
Yandex Webmaster site diagnostics: problems detected on the host (e.g. DNS errors, slow response, robots.txt issues, 4xx alerts), each with a severity (FATAL, CRITICAL, POSSIBLE_PROBLEM, RECOMMENDATION). By default returns only ACTIVE problems (state=PRESENT), worst first. Set includeAbsent=true to also list resolved/absent problem types. Read-only.
Inbound (external) links pointing to the host — the backlinks Yandex knows. Read-only. • report="samples" (default): example links with source page, target URL and discovery date, plus the total count available. • report="history": time series of the total external-link count over time.
Discovery tool: list the sites (hosts) available to your Yandex Webmaster token, each with its host_id, URL and verification state. host_id is what every other tool needs to identify a site. Optionally pass `hostId` to also fetch that host summary (SQI, indexed/excluded page counts, site problems). Read-only. Call this first.
How Yandex crawls and indexes the host. Read-only. Pick a report: • report="history" (default): time series of crawled pages grouped by HTTP status class (HTTP_2XX/3XX/4XX/5XX/OTHER) — spot spikes of errors the crawler hits. • report="crawled": example crawled URLs with their HTTP code and crawl date. • report="in_search": example URLs currently present in Yandex search, with title.
4 of 8 tools lack visible source code, only file paths provided. Cannot verify descriptions, schemas, or parameter details for get_search_queries, list_sitemaps, get_recrawl_status, submit_recrawl.
Output schemas not documented in code. Tools return structured results but LLMs cannot see field names, types, or required fields. Prevents downstream tool chaining and forces LLMs to infer structure.
No dry-run or confirmation pattern for submit_recrawl (WRITE operation). Agents can submit URLs for recrawl without preview or confirmation, risking unintended crawl load.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Error handling is generic. No recovery guidance in tool descriptions. If a tool fails, LLM has no hint whether to retry, call a different tool, or ask the user.
Tool descriptions do not explain when to call each vs. alternatives. E.g., get_indexing has 3 report types but no guidance on which to use for common tasks (spot crawl errors vs. find indexed URLs).