城市圖譜｜城市總覽 MCP server

• Registry nametw.cityatlas/cityatlasverified in the MCP registry

The registry name sits under cityatlas.tw. The registry only lets a publisher use that namespace after proving control of the domain through DNS or HTTP.

• Stateanswerschecked 2026-10-27
• Tools10
• Server calls itselfcityatlas
• Server version0.4.0
• Protocol version2025-06-18

On 2026-10-27 the server answered initialize and tools/list without a sign-in. First answer in 167 ms.

• Its tools do none of the product's use cases we checked: view a walkable city as a live 3D model; study urban canopy coverage in 3D.

Said by rules from the measurements above. The likely first call is a model's pick from the tool list; read its description before relying on it.

• INITE MCP score72 of 100answers 100 · does the job 0 · definitions 100 · context 100 · connects 100
• What an agent can do with it0 of 2 use cases
• Answered100% of days9 of 9 daily checks
• Median first answer329 ms
• Space in an agent's context≈ 2,001 tokensthe definitions of all its tools (10), which an agent carries on every turn
• Tool definitionsA · 100 of 100
• Typed results0 of 10 toolsdeclared outputSchema: results come typed, read once

• Every tool passes all six checks.

A program checks the server once a day: initialize and tools/list, no model. Tokens are the definitions' length at about four characters a token. The grade averages six checks of every tool's name, description, schema and annotations; it says how well the tools are described, not whether they work. The score weighs what was measured and nothing else: answering (25%), doing the product's job (25%), definitions (20%), context (10%), use (10%) and how readily a client connects (10%). A part that was not measured is left out, and a hidden instruction caps the score at 40. The state of vendors' MCP servers: method and numbers →

• search_catalog

  用關鍵字找城市圖譜的資料:11 大類 82 項資料,細到 161 條指標(141 條格網:事故、公車、人行道、淹水、樹…;20 條里 / 鄉鎮 / 縣市:里人口、長者與兒童、里所得…),加上 47 屆選舉與公投(投票率、得票率);公開版少 Google 街景判讀的 2 項 4 條,實際查得到 80 項、157 條:回代號、名稱、單位、年份、出處機關、是不是推估、涵蓋幾區。Search the indicator catalog by keyword (82…

• describe_metric

  一條資料的完整出處(機關、資料集、網址、取得日、檔案、對帳)、涵蓋哪些鄉鎮、統計、以及「可能在講別的事」的但書。Full provenance and caveats of one indicator.

• rank

  某條資料在里或鄉鎮排前幾名(跟網站 07 同一套算法:先加總再除、人少的地方往平均拉)。可換算每千居民 / 每公里道路 / 路 + 人 + 店(只有事故)/ 任何能當分母的代號;可只看某個鄉鎮、可加篩條件。只回前 N 名。里層資料(長者占比、里所得、選舉投票率…)一個里一個數,不能換算、單位只能是里(人數可加到鄉鎮)。沒編里的範圍不進排名。Rank neighbourhoods or districts by an indicator, optionally…

• lookup_place

  給地址(例:竹北市光明六路 10 號)、里名(例:竹北市東平里)或鄉鎮名 → 那裡各項數字,以及在大新竹排第幾;查里名時另附里人口、所得、65 歲以上與 14 歲以下。地址只定位到 125 公尺的格(用我們自己的門牌檔比對,不送外部服務、不記錄)。Numbers for an address, neighbourhood or district.

• need_ranking

  哪裡最多人被公車耽誤(被耽誤的人·分鐘、工作地、學校、公設的百分位加權);權重可帶,不帶用預設 60/20/10/10。Where the most people lose time on public transport compared with driving.

• overlay_metrics

  兩條資料疊起來看:A(可換算)與 B 各切三等分成九宮格,回各格有幾個里 / 鄉鎮,並列出「兩樣都高」的。例:事故(每千居民)高、人行道少的里。Which places are high on two indicators at once.

• compare_groups

  有 B 的格 vs 沒有 B 的格,A 差多少 —— 在居民 × 道路長度切成的 25 層裡各比一次(避免被市中心拉著走)。只描述,不做因果。Compare cells with vs without B, within strata of residents and road length.

• filter_cells

  找出同時符合最多四個條件的 125 公尺格,回符合幾格、住了多少人,並照里 / 鄉鎮加總。例:離公車站超過 500 公尺、又住超過 100 人。Find grid cells that meet up to four conditions.

• compare_change

  兩個年份比,哪些里增加最多、減少最多(跟網站 07「看變化」同一套):里人口(97–114 年底,戶籍;97–106 只有新竹縣市、竹南、頭份,新設里設立前是估計、不拿來比)、事故場數(111–114)、里所得中位數(111→112)、投票率(總統 / 立委不分區政黨票 / 立委區域 / 縣市長,年份照選舉目錄)。門檻自己填(增加幾人 / 件 / 萬元 / 個百分點以上才算增加很多);里界調整過、兩屆不是同一塊地的里不比、列出原因。只描述差多少,不做因果。Which…

• show_map

  把一條資料(或「需要度」)畫成里 / 鄉鎮的五級地圖,在對話裡直接顯示,可拖曳縮放、滑過看數字;換算同「排名」。畫不出來的 AI 會拿到一個能點開的網址(請轉給使用者)。Draw an indicator as an interactive choropleth map of neighbourhoods; chats that cannot render it get a link that opens the map in a browser (pass it on…