Exposes uax (User-Agent parser) over the Model Context Protocol. Identifies clients behind User-Agent strings: browser, engine, operating system, device, bot, in-app WebView, automation framework or HTTP library.
Four well-named, verb-prefixed tools with comprehensive descriptions (avg 180 chars, within baseline 194). All tools have input schemas with proper JSON Schema structure. Descriptions are LLM-optimized and include WHEN to use each tool and what it returns. However, output schemas are not documented, LLMs cannot see what fields to expect from responses. Error handling and recovery guidance are absent from visible code. No tool annotations (readOnlyHint, destructiveHint) despite all being read-only operations. Parameter descriptions are present and clear.
Check whether a client's claims contradict each other, and get a trust/risk verdict with the reasoning behind it. Detects impossibilities such as Safari on Windows or a non-WebKit engine on iOS. Use when deciding whether to believe a client. Findings are evidence of an edited string, not proof of intent, and the recommended action is advice — never an authorisation decision.
Report what this server can recognise: rule counts per category, the rule database fingerprint, and the device catalogue's size and snapshot date. Use to tell whether a negative answer means 'not a real client' or 'newer than the data', and to record which database produced a classification.
Resolve an Android model code such as SM-S928B or CPH2451 to its brand and marketing name. Use whenever a model code appears and the handset matters. A code the catalogue does not hold resolves its manufacturer from the code's shape and reports no name — it is never approximated, because a wrong device name is worse than no device name.
Identify the client behind a User-Agent string: browser, engine, operating system, device, bot, in-app WebView, automation framework or HTTP library. Every resolved field carries a confidence and the evidence it came from. Use this instead of reading a User-Agent yourself — the strings are deliberately misleading (every browser claims 'Mozilla/5.0'), and unresolved fields come back empty rather than guessed. Values frozen by Chrome's User-Agent Reduction are flagged as frozen rather than reported as observed facts.
Output schemas not documented. LLMs cannot see what fields parse_user_agent, check_consistency, lookup_device, and database_info return. This forces agents to guess at response structure and breaks downstream tool chaining.
No tool annotations. All four tools are read-only (Risk: READ_ONLY declared), but input schemas lack readOnlyHint annotation. This prevents agents from understanding which tools are safe to call speculatively vs. which require confirmation.
No error handling or recovery guidance visible in tool definitions. If parse_user_agent receives malformed input or database_info fails, LLMs have no guidance on what to do next or whether to retry.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
database_info accepts no parameters (empty props object). While correct for this use case, the description does not explain what triggers a call to this tool or when it is necessary relative to the other three tools.
lookup_device parameter 'code' description includes example 'SM-S928B' inline. Use enum constraints or a format pattern instead.