--- title: "How To Use Atlasly Inside Claude, ChatGPT, and Other AI Assistants" description: "Step-by-step guide to connecting Atlasly to Claude, ChatGPT, Perplexity and other AI tools so you can run UK site analysis, flood checks, and planning lookups directly inside the AI chat." canonical: https://atlasly.app/blog/how-to-use-atlasly-in-claude-chatgpt published: 2026-04-19 modified: 2026-04-19 primary_keyword: "how to use Atlasly in Claude ChatGPT" target_query: "how to use Atlasly in Claude or ChatGPT" intent: commercial --- # How To Use Atlasly Inside Claude, ChatGPT, and Other AI Assistants > Step-by-step guide to connecting Atlasly to Claude, ChatGPT, Perplexity and other AI tools so you can run UK site analysis, flood checks, and planning lookups directly inside the AI chat. ## Quick Answer To use Atlasly inside Claude or ChatGPT, open the AI's settings, find the Connectors or Custom Tools menu, add a new custom MCP server, and paste https://mcp.atlasly.app/mcp. Save and start a new conversation — asking "analyze 10 Downing Street" will return real UK site data (flood, heritage, planning, ground conditions) straight in the chat. ## Introduction Most architects already use Claude or ChatGPT for everyday research. The problem is those tools hallucinate on UK planning policy, can't see real flood zones, and definitely don't produce georeferenced CAD files. Atlasly's MCP connector fixes that. Once installed in your AI of choice, it gives the chatbot access to live UK planning data (NPPF, flood zones from the Environment Agency, listed buildings from Historic England, ground conditions from the British Geological Survey, and more) plus every tool from the full Atlasly platform — inside the conversation you're already having. This guide shows you exactly how to install it in Claude (the best experience today) and ChatGPT, and what to ask once it's connected. ## What you need before starting You need one of the following: - **Claude Desktop app** (free, Mac / Windows / Linux) — downloadable at claude.ai/download. Custom MCP connectors are supported out of the box. - **Claude.ai in a browser** — custom MCP connectors are available on Claude Pro, Team, and Enterprise accounts. - **ChatGPT Plus or Team** — custom MCP connectors are rolling out through OpenAI's custom tools framework. - **Any MCP-compatible client** — the server works with any client implementing the MCP standard (Cursor, Zed, Continue.dev, and others). You do **not** need: - An Atlasly account (public tools work anonymously). - An API key. - Technical knowledge of MCP protocols. You get better results with an Atlasly account — authenticated tools unlock project memory, site packages, and Site Watch monitoring. Sign up free at https://atlasly.app/auth. ## Installing Atlasly in Claude Desktop Claude Desktop has the smoothest install today. **Step 1.** Open Claude Desktop. If you have not installed it yet, download from https://claude.ai/download. **Step 2.** Click the **Claude** menu in the top-left → **Settings…** (or press ⌘ ,). **Step 3.** In the settings sidebar, click **Connectors**. **Step 4.** Under the **Web** section, click **Add custom connector** (or the **+** button). **Step 5.** Fill in: - **Name:** Atlasly - **URL:** `https://mcp.atlasly.app/mcp` **Step 6.** Click **Save**. Claude will connect to the Atlasly MCP server, read the tool manifest, and display 7 public tools (analyze_site, check_planning_constraints, get_flood_risk, get_planning_history, find_precedents, get_site_topography, check_walkability_transport) along with the Atlasly icon. **Step 7.** Start a new conversation. Ask something like "Using Atlasly, analyze the site at 10 Downing Street, London SW1A 2AA." Claude will call the `analyze_site` tool and return a real site breakdown including listed buildings, flood zone, and ground conditions. ## Installing Atlasly in ChatGPT ChatGPT's MCP support rolls out through the custom tools / connectors system on paid accounts. **Step 1.** Sign in to ChatGPT on a Plus, Team, or Enterprise account. **Step 2.** Open **Settings → Connectors** (the exact location shifts as OpenAI iterates; on some accounts it appears as **Custom GPTs → Custom Tools**). **Step 3.** Click **Add MCP server** (or the equivalent add-custom-tool button). **Step 4.** Enter the server URL: `https://mcp.atlasly.app/mcp`. **Step 5.** Save and start a new conversation. Ask the same question as above — ChatGPT should list Atlasly's tools in its available-tools menu and invoke them when the query matches. If your ChatGPT plan does not show the Connectors menu yet, the feature has not rolled out to your account. Claude Desktop is the most reliable option today. ## Authenticated workflows (optional) The 7 public tools work without an Atlasly account. To unlock the 4 authenticated tools — `create_project_from_site`, `run_site_package`, `generate_site_brief`, `enable_site_watch` — you authenticate once via OAuth. **Step 1.** Call an authenticated tool in your AI. For example, ask Claude "Using Atlasly, create a project for 10 Downing Street." **Step 2.** Claude opens a browser window pointing to https://mcp.atlasly.app/authorize. **Step 3.** You land on the Atlasly consent screen at https://atlasly.app/mcp/oauth/authorize. If you are not signed in, Clerk takes you through the Atlasly sign-in flow first. **Step 4.** You see exactly which permissions the AI is asking for — project writes, site package runs, Site Watch writes. Approve or deny. **Step 5.** Claude receives an access token and the four authenticated tools become callable. This is a one-time consent. Claude stores the refresh token so future conversations use the same authentication without re-prompting. ## Example prompts that work well The AI works out which Atlasly tool to use from natural language. You do not need to memorise tool names. Try these: - *"Using Atlasly, run a full site analysis for 10 Downing Street, London."* — triggers `analyze_site`. - *"Is 42 Berkeley Street, London W1J 5AW in a flood zone?"* — triggers `get_flood_risk`. - *"What are the planning constraints within 500 metres of Shoreditch Town Hall?"* — triggers `check_planning_constraints`. - *"Find planning precedents for mixed-use development near London Bridge."* — triggers `find_precedents`. - *"How walkable is 150 Aldersgate Street? Use a family persona."* — triggers `check_walkability_transport`. - *"Pull the last 10 planning applications near One New Change."* — triggers `get_planning_history`. Every response ends with a direct deep link into Atlasly where you can continue the analysis with maps, 3D context, and CAD/BIM exports. ## What the AI can and cannot do via the MCP **The AI can:** - Run first-pass site analysis using real UK government datasets. - Check flood zones, listed buildings, conservation areas, Article 4 directions, TPOs, environmental designations, historic landfill, and ground conditions. - Pull planning history from planning.data.gov.uk. - Screen walkability with persona-weighted amenity scoring and isochrones. - Return deep links into the Atlasly app with the site pre-loaded. **The AI cannot yet (via the MCP):** - Generate CAD/BIM exports directly — those are produced by the full Atlasly pipeline. The MCP provides a link to run them in the app. - Produce interactive 3D site models in the chat — those live in Atlasly's Site Studio. - Handle non-UK jurisdictions with the same depth — UK planning is production-first today; US / UAE / global coverage is partial and gracefully degrades to limited MCP responses. For anything the MCP cannot do, the returned payload always includes a deep link back into Atlasly so the workflow continues seamlessly. ## Troubleshooting **The connector shows up but tools fail.** Try removing and re-adding the connector. Claude caches MCP metadata; a fresh install forces a re-handshake. **Claude shows "signed out" on the consent page.** Atlasly uses Clerk for authentication. Clerk cookies are set at `.atlasly.app` scope so they cover `mcp.atlasly.app` too. If you see signed-out behaviour, sign in at https://atlasly.app first, then retry. **"Address not resolved" errors.** Use full UK postcodes and address strings. Try "10 Downing Street, London SW1A 2AA" rather than just "Downing Street". **Non-UK site queries return limited data.** Intentional — UK is production-first. Open the returned Atlasly link for the full interactive workflow globally. **The ChatGPT Connectors menu is missing.** The feature has not rolled out to your account yet. Use Claude Desktop in the meantime. ## From Practice We had a practice ask whether the MCP actually saved time in real workflows. Their test: run the first-pass constraints for a Shoreditch site live in a Monday team meeting. In 18 seconds Claude was back with flood zone, 12 listed buildings, BGS Class 3 ground conditions, and 34 nearby planning applications — all as clickable links. That used to be an intern's afternoon. The rest of the meeting was spent actually discussing what to do with the constraints, which is what the hour was meant to be about. ## Frequently Asked Questions **Do I need to install anything to use Atlasly in Claude?** No. You paste a URL (https://mcp.atlasly.app/mcp) into Claude's Connectors settings and Claude handles everything else. No downloads, no npm installs, no config files. **Is the Atlasly MCP really free?** Yes. The 7 public tools work anonymously with no account required. The 4 authenticated tools require a free Atlasly account for OAuth but the account itself is free. **Does the Atlasly MCP work in ChatGPT?** Yes on paid ChatGPT plans that have the Connectors / custom tools feature enabled. OpenAI's rollout schedule varies; Claude Desktop is the most reliable option today. **What MCP clients are supported besides Claude and ChatGPT?** Any MCP-compatible client works. Tested with Claude Desktop, Claude.ai web, ChatGPT, Cursor, Zed, Continue.dev, and Perplexity's custom-tools feature. The server follows the Model Context Protocol specification strictly so new clients should work without changes. **Does the Atlasly MCP export DXF or DWG files directly into the chat?** The MCP returns a signed link to the CAD export hosted on atlasly.app — the actual file is produced by the full Atlasly pipeline. One click from Claude opens the file in AutoCAD or Revit. This keeps chat responses fast and lets the heavy geometry work happen in the Atlasly runtime. **Which countries does the Atlasly MCP cover?** UK is production-first — planning data, heritage, ecology, flood, and ground conditions all use live UK government datasets. US, UAE, and global sites return limited first-pass data plus a deep link into the full Atlasly app for interactive review. **Can my firm deploy the Atlasly MCP on a private server?** The hosted MCP at mcp.atlasly.app is the recommended install for most firms. For regulated or on-premise deployments, contact us via https://atlasly.app/pilot — Enterprise plans support private deployment options. ## Conclusion Claude Desktop with the Atlasly MCP installed is the fastest way an architect can move from "I have an address" to "I have real UK planning intelligence." Install once, use it across every project. Try it free — paste https://mcp.atlasly.app/mcp into your AI's Connectors settings and ask about any UK site. Every response ends with a link into the full Atlasly app for maps, 3D context, CAD/BIM exports, and client-ready PDF reports. ## Related Reading - https://atlasly.app/blog/atlasly-vs-planningbot - https://atlasly.app/blog/atlas-ai-free-architecture-planning-assistant - https://atlasly.app/blog/export-dxf-dwg-site-analysis-from-ai-chat - https://atlasly.app/blog/ask-ai-for-uk-planning-constraints-nppf --- Source: https://atlasly.app/blog/how-to-use-atlasly-in-claude-chatgpt Platform: Atlasly — AI site intelligence for architects, engineers, and urban planners. https://atlasly.app