Tools
This table is here so you can judge whether ArchLink MCP covers enough to be worth wiring up.
Each row is one thing your AI client can go and do once it is connected. You never have to remember these names — you just describe what you want. The table is for the decision you make before connecting: how far the coverage reaches, and how long each thing takes.
The duration column is deliberate. A lot of early-stage architectural work is slow by nature: it means searching the live web, collecting data about a site’s surroundings, or rendering maps and charts. Saying so up front lets you tell apart the things you can ask mid-conversation from the things you should kick off and then go do something else.
This table cannot go stale
The names, the return mode and the duration ranges are not hand-written. They are generated from the running service’s own tool registry, and CI fails outright if the generated table and the service disagree. If you can see this table, it is current.
What you read here is not what your client receives
The one-line summaries in this table were written for this page, in English.
The tool descriptions the server actually sends to your client are a different text altogether: they are written in Chinese, and they are instructions aimed at the model — how to call the tool, what to pass, what to double-check with you — rather than a description aimed at a reader. The two are deliberately not translations of each other.
Worth knowing before you connect: your client will show you those Chinese descriptions.
| Tool | What you get | How it returns | Takes |
|---|---|---|---|
search_standards | The text of the Chinese national and industry building codes, quoted rather than paraphrased by a model. | Straight away | Seconds |
search_cases | A shortlist of reference projects with name, location, year, floor area, category, cover image and source link. | Straight away | Seconds |
get_case_detail | Everything held on one project, including its full image set. | Straight away | Seconds |
locate_site | Turns a spoken address or landmark name into coordinates, so nobody has to look them up on a map first. | Straight away | Seconds |
expand_prompt | Turns one sentence into a set of architectural image prompts, phrased in the vocabulary render engines respond to. | Straight away | Seconds |
analyze_policy | A written read of the policy environment for a site, traced down the six-level Chinese planning hierarchy, plus the tier above it. | As a job | ~2–10 min |
analyze_economy | A city's economic fundamentals — GDP, population, urbanisation, industry mix, retail, household income — as charts and data tables with a written read; any chart the data can't support is flagged, not faked. | As a job | ~5–25 min |
recommend_reference_cases | Picks up to eight reference projects from the case library for your building type and brief, and distils a set of design strategies you can borrow. | As a job | ~2–10 min |
analyze_planning_condition | What a specific plot may be built to: retrieved regulatory-plan limits and design guidelines, with plot ratio worked out from the areas you give. | As a job | ~2–10 min |
analyze_culture_context | The cultural and historical threads of a district, condensed into cues you can actually take to a facade. | As a job | ~2–10 min |
analyze_market_product | Competing schemes and market benchmarks nearby, read against your project's scale to suggest who it is for. | As a job | ~2–10 min |
analyze_location | An overall location read — nearby amenities, entrance orientation, display frontage, road-network logic — with location drawings; a site boundary is best, but a single point works too. | As a job | ~3–20 min |
A job returns an id immediately; you collect the result later at the interval it tells you, instead of holding the conversation open.