Placepoint MCP

Norwegian property and company data in your AI assistant: owners, tenants, sales, land register, zoning plans and maps, with links back to Placepoint Fusion.

View on placepoint.no

Hosted MCP Server

npx add-mcp 'https://mcp.placepoint.no'

Installs into Claude Code, Codex, Cursor and more

Documentation

Placepoint AI makes Placepoint available in Claude: in Claude Desktop, Claude Code and Cowork. Three MCP servers give Claude all of Placepoint, Norway and Sweden, both data and features, and you write what you want to know in the language you use yourself. The servers also work in Microsoft Copilot Studio and ChatGPT.

Claude retrieves Placepoint's combined, contextualised and quality-checked data (matrikkel (the Norwegian property register), Grunnboken (the land register), company data, map layers and more, see Data and sources), combines it with open sources and your own datasets, and puts it all together into an answer, a table, a map, an interactive dashboard or a PDF report of the same type as the reports in Placepoint Fusion.

There are three servers. Placepoint MCP gives Claude the registers and the map layers for Norway. Placepoint Dataset MCP gives Claude your projects and datasets in Placepoint Fusion, so the result of an analysis can become a map layer you keep working on, and your lists can meet the registers in the same question. Combify MCP gives Claude Swedish property. What each of them provides is set out in Three servers, one conversation.

[!-info] -info info

Placepoint AI is under rapid development. New tools and data sources arrive continuously, and features, data and access can change along the way. What is new is on Updates.

What is MCP?

MCP (Model Context Protocol) is an open standard for giving an AI assistant access to tools and data sources, and it is used by Claude and a range of other AI tools. When Placepoint AI is connected in Claude, Claude can look things up in our property registers while the two of you talk, instead of guessing from what the model remembers. You do not need to know anything about matrikkel structure, field names or APIs. You ask, Claude retrieves.

The foundation is the same as in Placepoint: around 2,700 sub-layers from public agencies, municipalities and commercial vendors, covering 3.5 million matrikkelenheter (cadastral units), 1.1 million active companies and more than 30 million ownership relationships between companies, owners and property. Each individual dataset, what it contains and where you find it at the source, is listed under Data and sources. The scope in numbers is on Placepoint in numbers.

Three servers, one conversation

Placepoint MCP reads the Norwegian registers and map layers. Placepoint Dataset MCP reads and writes your own projects and datasets. Combify MCP reads the Swedish registers.

Placepoint MCPPlacepoint Dataset MCPCombify MCP What Claude gets The registers and the map layers: matrikkel, Grunnboken, companies, land-use plans, risk, travel time Your projects and datasets in Placepoint Fusion Swedish property: fastigheter, bygglov, detaljplaner, markanvisninger and companies Direction Reads. Nothing you ask changes anything in Placepoint Reads and writes. Claude can create datasets, add rows and columns, and change the style of the map layer Reads. Nothing you ask changes anything in Combify Typical question "Who owns [address], and what is registered on the property?" "Upload our tenant list, link it to the matrikkel and show it on the map." "Who owns Solna Svärdsliljan 1, and which bygglov have been granted on the fastighet?" The result An answer, table, map, dashboard or PDF report in Claude A dataset in your project, visible and editable in the map An answer, table, map, dashboard or PDF report in Claude, and extracts to Excel or CSV Sign-in Your Placepoint user, at https://mcp.placepoint.no Your Placepoint user, at https://mcp-dataset.placepoint.no. Access follows the projects you have access to Your Placepoint user. You get the URL for the server from your contact at Placepoint
What Claude getsThe registers and the map layers: matrikkel, Grunnboken, companies, land-use plans, risk, travel timeYour projects and datasets in Placepoint FusionSwedish property: fastigheter, bygglov, detaljplaner, markanvisninger and companies
DirectionReads. Nothing you ask changes anything in PlacepointReads and writes. Claude can create datasets, add rows and columns, and change the style of the map layerReads. Nothing you ask changes anything in Combify
Typical question"Who owns [address], and what is registered on the property?""Upload our tenant list, link it to the matrikkel and show it on the map.""Who owns Solna Svärdsliljan 1, and which bygglov have been granted on the fastighet?"
The resultAn answer, table, map, dashboard or PDF report in ClaudeA dataset in your project, visible and editable in the mapAn answer, table, map, dashboard or PDF report in Claude, and extracts to Excel or CSV
Sign-inYour Placepoint user, at https://mcp.placepoint.noYour Placepoint user, at https://mcp-dataset.placepoint.no. Access follows the projects you have access toYour Placepoint user. You get the URL for the server from your contact at Placepoint

Together they cover the workflow from question to delivery. Claude finds the properties with Placepoint MCP, and saves the result as a dataset with Placepoint Dataset MCP. The next time you open the project in Placepoint Fusion, the list is there as a map layer. In the other direction, you upload your own list, a portfolio, a customer list or a tenant register, and ask about it together with the registers: which properties on the list have a new owner or a new charge since last time? If you have property in both countries, Claude answers across them in the same conversation.

From days to minutes

Analysis work around property is largely manual: find the right register, extract the data, quality-check it and put it all together into something the board, the credit committee or the client can read. That takes days or weeks. With Placepoint AI, Claude does the same work in minutes, and delivers the result as a report, an interactive dashboard or a presentation. Some typical tasks:

The question does not have to concern a single property. Ask about a whole street, a block or a district, and Claude does the job for all the properties in the area at once. The prompt "Check the whole of John Colletts allé, where can new buildings go, produce a report" gave a 15-page report that went through all 72 land parcels in the street, with a map and an assessment per address. Along the way, Claude read the plan documents themselves, and found four addresses in the middle of the street that are not covered by the building ban that stops all the neighbours. That exemption is not in any map database. It is in the decision on the extended building ban, because the requirements are in the planning provisions, not in the colours on the plan map. Going through 72 properties and their plan documents by hand takes days; here it took minutes.

The foundation is the data we have already collected, quality-checked and put in context: see Data and sources. Claude retrieves from that same foundation, so you avoid searching each individual source and stitching it together yourself.

What you can do with Placepoint AI

The examples below are lookups, reports and analyses our users already do with Placepoint AI. Replace the placeholders in square brackets with a real address, a matrikkelnummer (cadastral number) or a company registration number:

With Placepoint MCP

  • Look up [address] or [matrikkelnummer] and gather facts about the site, buildings, units, owners and tenants in one place, the way the property panel and the units and tenants tab do it in Placepoint Fusion.
  • Search and filter properties on criteria across the whole of Norway, for example year of construction, building type, area or tenants, and export the result. The quick filters from Filter in Placepoint Fusion also work here, such as office, undeveloped, SPV-owned, sold in the last twelve months and energy rating A to C.
  • Search in areas the way people name them: grunnkrets (basic statistical unit), delområde (sub-area) and tettsted (urban settlement) are looked up by name, such as Ullevål Hageby or Frogner.
  • See all registered sales of a property, with buyer, seller, amount and date, or search for sales above an amount in a municipality.
  • Retrieve value estimates for homes from our home valuation model, with value, range and rent estimate per unit, and totals per property, housing cooperative or grunnkrets.
  • Retrieve the full land register extract for [matrikkelnummer] and summarise the encumbrances: charges, easements, rights of way and subdivision bans.
  • Bulk-scan Grunnboken records and encumbrances for a whole area at once, for example all matrikkelnummer within [area].
  • Map the companies behind [matrikkelnummer]: the whole ownership chain with effective ownership share and ultimate parent, roles, shareholders and annual accounts, as in the company panel. The general manager and the board are included for both the owners and the tenants.
  • Find and rank companies by industry, municipality, employees, turnover and whether they own property or have tenants. Branches are listed under their own name, so all the shops in a chain can be listed location by location.
  • Produce a due diligence report for [matrikkelnummer] with a legal, technical and financial review.
  • Produce a credit and risk report for [matrikkelnummer].
  • Produce a climate risk and nature risk report for [matrikkelnummer] for use in a credit assessment, with traffic lights (red, amber, green) per topic such as flooding, landslides, quick clay, surface water, cultural heritage and threatened species.
  • Produce a climate risk and sustainability analysis for a head office at [address], with tenants, certification and travel time.
  • Check the risk and the framework around [matrikkelnummer]: flooding, landslides, quick clay, zoning and municipality plans, habitat types and protected areas, shown on a map with the risk map layers.
  • Find out which land-use plans apply to [matrikkelnummer], how large a share of the site each of them covers and whether planning work is in progress, including municipality plan proposals. The plan that applies to the site is kept separate from neighbouring plans that only touch the boundary, and the plan map can be shown together with the property boundaries, in line with the zoning map layers.
  • Retrieve the plan documents for a plan, for example the planning provisions and the plan description, and have them read and summarised, not just listed.
  • Search for land-use plans in an area or by plan name, and see plan maps and plan proposals alongside what has been adopted, including detailed zoning plans and area zoning plans.
  • Retrieve historical aerial photos around [matrikkelnummer] and compare the area year by year.
  • Calculate travel time and reach by car, public transport, bicycle or on foot, and find properties within a given travel time. With public transport, you choose the departure time yourself.
  • Do a location analysis: the best place for a new shop in [area], or suitable sites for data centres in Norway.
  • Produce an investment and development analysis for [matrikkelnummer]: demolish or extend, yield, development density, development potential and possible land use.
  • Produce a buyer assessment of [company registration number] that wants to take over [matrikkelnummer].
  • Retrieve the full property portfolio of [company], including properties owned indirectly through subsidiaries and SPV chains.
  • Produce a portfolio or board report for [company registration number], of the same type as the reports in Placepoint Fusion.

With Placepoint Dataset MCP in addition

  • Save a search result as a dataset in your project, for example all logistics properties over 10,000 m² in [municipality] with owner and year of construction, and get it as a map layer in Placepoint Fusion.
  • Upload an Excel file and have the rows linked to the matrikkel by address or matrikkelnummer, the way dataset upload does it in the platform. If you have the geometry in GeoJSON or WKT, put it in a column in the spreadsheet.
  • Add municipality, grunnkrets or postcode as columns in a dataset, and sum or count rows per area.
  • Ask about your own rows together with the registers: which properties in [dataset] have changed owner, received a new charge or lie in a flood zone?
  • Calculate travel time from each point in a dataset, or find the shortest route that visits them all.
  • Tidy up a dataset: change values, add columns, hide columns and set colours and symbols on the map layer.

With Combify MCP

  • Fastigheter: a lookup by address or fastighetsbeteckning, such as "Solna Svärdsliljan 1", or a search in an area, with an owner or in a value range. Fastigheter can be filtered and averaged on owner type (municipality, company or private person), fastighetstype, taxeringsvärde split into land value and building value, assessed rental income and whether there is a linked plan. The owner's phone number is included where it is registered.
  • Bygglov: filter on the company behind the work and on whether the property is municipal, private or owned by a person. Rank the most active developers in a municipality, with subsidiaries gathered under the parent company.
  • Plans: detaljplaner, planbesked and markanvisninger, with the source document where it exists. Markanvisninger can be narrowed by area and price per m².
  • Companies: search by name, company registration number, industry and municipality, and the whole ownership structure up to the parent company and down through the subsidiaries, with the properties the group owns gathered together.
  • Everything connected to one thing, in a single lookup: if you ask about a fastighet, a bygglov, a detaljplan or a markanvisning, you get the other bygglov, plans, properties and companies linked to it.
  • Area statistics for an area, reports and summaries, and extracts to Excel or CSV.

Swedish specialist terms match the way you write them, and each of them is explained in the glossary. New fields and filters arrive continuously; what has been added is on Updates.

Finished reports, with a complete report example to download, and short example prompts you can start with, are on Examples. One of them is a demographics analysis for a catchment area around Lørenvangen 22, with the prompt and the finished PDF report; the manual route behind the same question is on Demographics analysis.

How reports are delivered

Reports are delivered as HTML by default, because that is fastest and lets you read the result straight away. Ask for PDF when you need a document to send on, and choose between one long page or genuinely paginated pages. PDF takes longer to produce, and Claude tells you how long it will take before it starts.

When a plan is discussed in a report or analysis, Claude reads the planning provisions and the plan description first, instead of building the text on the plan name alone. Property lookups also suggest plan status and a hazard sweep (landslides, flooding, quick clay and similar) as the next step, so zoning and risk are not skipped. The reports resemble the ones you produce in Analyses and reports.

Which data is available?

Claude retrieves from the same sources that underpin Placepoint, in Norway and in Sweden.

Norway

In total, Claude retrieves from more than 130 datasets delivered by around 25 organisations. Most are open, public sources, among them Kartverket, Brønnøysundregistrene, NVE, NGU, Miljødirektoratet, DIBK, Artsdatabanken, Statistisk sentralbyrå (SSB) and Statens vegvesen. In addition come commercial vendors and services, among them Geodata, Microsoft Azure Maps (background maps and places) and TravelTime (travel time).

You will find an overview on Data and sources, and who delivers what on Vendors.

Sweden

We cover Swedish property through Combify, which we bought in July 2026. Combify MCP gives Claude and the Agent the Swedish data: 4.2 million fastigheter in all 290 municipalities, more than 500,000 bygglov and close to 120,000 detaljplaner. The sources are Lantmäteriet, Skatteverket, the municipalities and Bolagsverket. The scope is on Placepoint in numbers.

Can be combined with other systems

Placepoint AI is not meant to replace the specialist systems you already use, but to work alongside them. The property data Claude retrieves can be combined with information from other solutions, for example systems for physical property management (leases, areas, operation and maintenance), ERP and finance systems, and specialist systems for the management, operation and maintenance of public building stock. Then you can use Placepoint for matrikkel, Grunnboken, company and risk data, and your own systems for what they are best at, in the same analysis or report.

In Claude Desktop, in Cowork and in Claude Code you connect to other MCP servers yourself, alongside Placepoint's. In the Agent in Placepoint Fusion, the administrator sets up the connections to external systems; which systems are ready and which are coming is described on Placepoint Agent.

The Agent in Placepoint Fusion

Placepoint AI is also available inside Placepoint Fusion, as the Agent, in pilot. There you ask the same questions directly in the platform, without Claude Desktop, Cowork or Claude Code. The Agent uses all three MCP servers, controls ordinary Fusion features such as the map, the map layers and the property panel, searches this user documentation when you wonder how something is done, and can be connected to other systems the business uses. The Agent is being trialled with selected users. It produces neither images nor files, so historical aerial photos, density maps and export to spreadsheets are only available in Claude; in the Agent the answer comes as text and map links.

Personal data is not available

Placepoint AI does not give out information about private individuals who own property. When Claude retrieves a land register extract, owners or transactions, personal names and national identity numbers are removed before the data reaches the model, and private individuals are shown as Private person. So you cannot use Placepoint AI to find out who owns a home. Companies, roles and ownership structure are open data and are included. This also covers the names and birth years of private individuals who hold roles in Brønnøysundregistrene or are shareholders in Aksjonærregisteret (the shareholder register), since both registers are public. National identity numbers are never provided.

This applies regardless of which access level your account has elsewhere in Placepoint. The extended access to personal data that some professional users can get in the platform itself (documented "legitimate interest") does not apply to Placepoint AI.

Placepoint Dataset MCP works with your own datasets, and only sees the projects your Placepoint user has access to. The content of the files you upload is your responsibility in the same way as when you upload them in the platform.

Who is the data controller when your AI tool retrieves data from us, and when you need a data processing agreement for Dataset MCP: Privacy and data processing.

How to install

[!-info] -info info

Placepoint AI requires a separate licence. What it provides and how you get it is on Placepoint AI at placepoint.no.

Placepoint AI is developed for and tested with Claude from Anthropic. The servers follow the MCP standard and also work in Microsoft Copilot Studio, OpenAI ChatGPT, Codex, Cursor, Perplexity, Gemini Enterprise and Gemini CLI, but Claude is the one we invest in and recommend: that is where we test every new feature first, and that is where the tools give the best answers.

Each server is added as a separate MCP server, with the same URL and the same sign-in wherever you connect. You sign in once per server, with the same user you use in Placepoint Fusion, and access follows the user, not the URL.

ServerURLGives access to Placepoint MCP https://mcp.placepoint.no The registers and the map layers Placepoint Dataset MCP https://mcp-dataset.placepoint.no Your projects and datasets, the same ones you see in Placepoint Fusion Combify MCP You get it from us Swedish property, see Sweden
Placepoint MCPhttps://mcp.placepoint.noThe registers and the map layers
Placepoint Dataset MCPhttps://mcp-dataset.placepoint.noYour projects and datasets, the same ones you see in Placepoint Fusion
Combify MCPYou get it from usSwedish property, see Sweden

Which account do I sign in with?

The same one as in Placepoint Fusion. The servers have no users or keys of their own: when Claude asks you to sign in, you use the account you sign in with at app.placepoint.no, and access follows that account. One exception applies regardless of account: personal data is never available.

Each person signs in themselves, also when an owner has added the servers for the whole organisation in Claude. That applies whatever the owner chooses in the dialogue that appears when the server is added: Sign in now signs the owner in straight away, Sign in when needed waits until the owner uses the server for the first time. Both work. Choose Sign in now, then you can see that the sign-in works before you ask colleagues to connect. They find the server under Connectors, click Connect and sign in with their own Placepoint account.

Never share one Placepoint account between several people. Who gets to see what is assessed per account, see Account and sign-in. If a colleague does not have a user in Placepoint, the administrator adds a new one in the usual way.

Who has AI access?

The account needs a seat on your organisation's licence for Placepoint AI. We set up the licence together with you, and it covers either everyone in the organisation or named people. The same seat gives answers from Placepoint data in the Agent. If you do not have a seat, ask the administrator in your organisation or contact us.

Claude Desktop, Claude on the web and Cowork

The procedure is the same in Claude Desktop, in Claude on the web and in Cowork, and it is described in Get started with custom connectors using remote MCP. Your own MCP servers are available on Pro, Max, Team and Enterprise, and on Free with one server.

  1. Open Settings and select Connectors. On the web it is under Customize.
  2. Click Add custom connector, paste the URL and click Add: https://mcp.placepoint.no for Placepoint MCP, https://mcp-dataset.placepoint.no for Placepoint Dataset MCP.
  3. Click Connect next to the server and sign in with your Placepoint user.
  4. Repeat for the other servers.

On Team and Enterprise, an owner adds the servers for the whole organisation under Organization settings, Connectors, with Add, Custom and Web. Each user then signs in themselves under Connectors, see Which account do I sign in with?.

If new tools are missing from the list, go to Connectors, select Placepoint, click the three dots and select Refresh tools list.

If you get the message "You signed in to Placepoint, but the server then rejected the request from Claude", the sign-in went through, but the server did not give the account access. This happens when the account does not exist in Placepoint Fusion, is blocked or lacks a licence for Placepoint AI. Disconnect the server under Connectors, connect again and sign in with the account you use at app.placepoint.no, see Which account do I sign in with?. If the message appears again, write to us at support@placepoint.no with the email address you signed in with, and we will check your access.

Claude Code

In Claude Code you add the servers from the terminal, one command per server, each with its own name:

claude mcp add --transport http placepoint https://mcp.placepoint.no
claude mcp add --transport http placepoint-dataset https://mcp-dataset.placepoint.no
claude mcp add --transport http combify <URL-en til Combify MCP>

Then type /mcp in Claude Code, select the server and sign in in the browser that opens. The options, for example whether the server should apply only to you or to the whole project, are described in Connect Claude Code to tools via MCP.

Microsoft Copilot Studio

Placepoint AI can be used from an agent you build in Microsoft Copilot Studio, and the agent can be published to Microsoft 365 Copilot and Teams. That requires a Copilot Studio licence in your Microsoft 365 environment, see Licensing for Copilot Studio. The chat in Microsoft 365 Copilot cannot connect to MCP servers directly, so it is through an agent that the servers become available.

The procedure is described in Connect your agent to an existing MCP server. The agent must be built on what Microsoft calls the standard harness; on the front page in Copilot Studio you select Other ways to build and Standard agent.

  1. Open the agent, go to Tools and select Add a tool, New tool and Model Context Protocol.
  2. Fill in Server name, Server description and Server URL: https://mcp.placepoint.no for Placepoint MCP, https://mcp-dataset.placepoint.no for Placepoint Dataset MCP. The agent uses the description to decide when the server should be called, so write what it provides, for example "Property, company and map data from Placepoint".
  3. Select OAuth 2.0 as the authentication and Dynamic discovery as the type. The server provides the sign-in endpoints itself, so no other fields need to be filled in.
  4. Click Create and then Next.
  5. Under Add tool, select Create a new connection, sign in with your Placepoint user and click Add to agent.
  6. Repeat for the other servers.

Write a short instruction under Instructions on the agent's Overview page, so the agent knows when to use Placepoint AI. For example: "Use Placepoint AI for all questions about properties, addresses, companies, zoning and maps in Norway. Call get_usage_guide first in every conversation." See Configure high-quality instructions for generative orchestration.

Select the tools the agent needs

An agent in Copilot Studio can have at most 128 tools switched on, in total across all the servers it is connected to. Placepoint MCP has around 100 tools and Placepoint Dataset MCP around 70, so with all the tools switched on in both there are too many, and the agent replies with an error message. Microsoft also recommends 25 to 30 tools per agent so that it chooses the right tool, see Add tools to custom agents.

Open the server under Tools, turn off Allow all and turn on the tools the agent needs. Leave get_usage_guide and get_tool_help on, because they tell the agent how to use the other tools. Tools we add later are switched off until you switch them on. The table shows which tools in Placepoint MCP belong to the most common tasks.

TaskTools Property and owner search, get_document, search_cadastre_by_address, count, aggregate Zoning and plans get_property_plans, search_plans, get_plan, get_plan_documents, fetch_plan_document, get_plan_dossier, render_plan_map Company search_companies, get_company_document, lookup_company_brreg, company_dossier, get_company_shareholders, get_company_roles Land register and sales get_grunnbokutskrift, get_property_transactions Map layers and risk search_map_layers, map_layer_lookup, query_map_layer, map_layer_risk_sweep
Property and ownersearch, get_document, search_cadastre_by_address, count, aggregate
Zoning and plansget_property_plans, search_plans, get_plan, get_plan_documents, fetch_plan_document, get_plan_dossier, render_plan_map
Companysearch_companies, get_company_document, lookup_company_brreg, company_dossier, get_company_shareholders, get_company_roles
Land register and salesget_grunnbokutskrift, get_property_transactions
Map layers and risksearch_map_layers, map_layer_lookup, query_map_layer, map_layer_risk_sweep

If the agent needs more tools than there is room for, you have two options. You can connect only the servers the agent uses: Placepoint Dataset MCP is only needed when the agent is to read or change your own datasets in Placepoint Fusion. Or you can split the work across several agents, because each child agent has its own limit of 128 tools, see Add other agents.

Test the agent in the test panel and publish it to Microsoft 365 Copilot or Teams, see Publish your agent. Each user of the agent signs in with their own Placepoint user the first time.

MCP servers go through Power Platform connectors, so a data policy in the Power Platform Admin Center applies to them too. If the server is blocked, the administrator must allow it in the policy.

If your organisation has turned on tenant isolation in Power Platform (Restrict cross-tenant connections), sign-in stops with an error message about "tenant isolation policy". This happens because you sign in to our Entra tenant, not your organisation's. A Power Platform administrator allows it in the Power Platform Admin Center under Security, Identity and access and Tenant isolation: click Add exceptions, select the Outbound direction and enter the tenant ID db862da0-eaab-41e5-8b86-48fd893b8d63. See Cross-tenant inbound and outbound restrictions.

If you get "Can't create MCP server. Try again." when you click Add, there is usually already a tool or a connector with the same name in the environment, for example from an earlier attempt. Use a new name, or delete the old one under Tools in the agent and under Custom connectors at make.powerapps.com.

If you get the error message "Tools cannot have more than 128 elements" (ToolLimitExceeded), the agent has more than 128 tools switched on in total. Switch off tools as described under Select the tools the agent needs. If the agent says it is missing a tool, for example get_property_plans, the tool is switched off. Switch it on under Tools.

If the agent answers "No information was found that could help answer this" without calling any tools, it is using classic orchestration, and then it never sees the MCP servers. This happens with agents that were created before the Copilot Studio licence was active. Open Settings, select Generative AI, set Orchestration to Yes and save. Then start a new test session.

If the agent answers "Eskalering til en representant er for øyeblikket ikke konfigurert" (escalation to a representative is not currently configured) without calling any tools, it found no tools that matched the question. Open the server under Tools and check whether the tool list is empty. If it is, select Create a new connection and sign in again, or remove the server and add it again. If you do not have a Placepoint AI licence, the agent replies with what is missing and who can give you access.

OpenAI ChatGPT

Placepoint AI can be used in ChatGPT on the web, not in the mobile apps. You add the servers under Plugins, which is described in Connecting custom MCP servers in ChatGPT. What you get depends on your subscription: on Business, Enterprise and Edu all the tools work, on Pro only the ones that read. The tools in Placepoint Dataset MCP that create and change datasets therefore require Business, Enterprise or Edu.

On Business, only administrators and owners can add the servers. On Enterprise and Edu, users the administrator has given the Create plugins with MCPs permission can also do it.

  1. Go to Plugins, click the plus button and select Add custom MCP server.
  2. Type a name, for example Placepoint, and a short description of what the server provides.
  3. Paste the URL under Connection: https://mcp.placepoint.no for Placepoint MCP, https://mcp-dataset.placepoint.no for Placepoint Dataset MCP.
  4. Select OAuth as the authentication, tick I understand and want to continue and click Create as a plugin.
  5. Sign in with your Placepoint user when ChatGPT asks you to.
  6. Repeat for the other servers.

The URL must start with https://. If it says http://, the server redirects the request, and the connection fails without an error message.

If you do not see Plugins, your workspace still has the older setup. Then go to Settings, Apps and click Create, fill in the same details, click Scan Tools and then Create.

Ask ChatGPT to use Placepoint

Type @ in the message field and select Placepoint. The choice applies only to that one message, so mention Placepoint again when a follow-up question needs new data.

Without a standing instruction, ChatGPT often searches the web or downloads documents from the municipality before it asks Placepoint. Open Settings and Personalization, check that Enable customization is turned on, and paste something like this into Custom Instructions:

Use Placepoint for all questions about properties, addresses, companies,
zoning and maps in Norway, before you search the web or download
documents. Call get_usage_guide first in every conversation. Fetch documents
from the municipality's public records portal only when Placepoint does not
have the answer, and say so when you do.

The instruction applies to all your conversations. More advice on instructions is in Give Claude instructions once, and it applies just as well to ChatGPT.

Publish for the whole workspace

An administrator or owner tests the server and publishes it for everyone:

  1. Open Admin Console, select the workspace, open Plugins and select Manage Legacy Apps.
  2. Select Drafts and Publish. On Enterprise and Edu you also select which groups get access, under Configure Access, and which tools are allowed, under Configure Actions.

Users then find Placepoint under Settings, Plugins and sign in with their own Placepoint user.

ChatGPT freezes the list of tools when the server is published. When we add new tools, an administrator opens the server under Plugins and clicks Refresh.

OpenAI Codex

Codex connects to your own MCP servers from the terminal, in the ChatGPT app on Mac and Windows and in the IDE extension, see MCP servers. From the terminal, two commands per server:

codex mcp add placepoint --url https://mcp.placepoint.no
codex mcp login placepoint

In the app and the IDE extension, open Settings, select MCP servers and Add server, select Streamable HTTP and paste the URL: https://mcp.placepoint.no for Placepoint MCP, https://mcp-dataset.placepoint.no for Placepoint Dataset MCP. Save, click Authenticate and sign in with your Placepoint user.

Cursor

Cursor connects from the desktop app, see Model Context Protocol. With Cursor installed, you add a server with one click: Placepoint MCP and Placepoint Dataset MCP. You can also add the server yourself to ~/.cursor/mcp.json, or to .cursor/mcp.json in the project:

{
  "mcpServers": {
    "placepoint": {
      "url": "https://mcp.placepoint.no"
    }
  }
}

Then open Cursor Settings, select MCP, click the server and sign in in the browser that opens. Agents that run at Cursor in the cloud cannot sign in yet.

VS Code with GitHub Copilot

VS Code adds a server with one click when VS Code is installed: Placepoint MCP and Placepoint Dataset MCP. Confirm with Install, start the server and sign in with your Placepoint user when VS Code asks you to. You can also select MCP: Add Server in the command palette, select HTTP and paste the URL. We have not tested VS Code ourselves.

Perplexity

Perplexity connects to your own MCP servers as custom connectors on the web, see Adding Custom Remote Connectors. It requires Pro, Max or Enterprise.

  1. Open Settings and select Connectors.
  2. Click + Custom connector and select Remote.
  3. Fill in the name, the URL and a short description of what the server provides. The URL is https://mcp.placepoint.no for Placepoint MCP, https://mcp-dataset.placepoint.no for Placepoint Dataset MCP. Select OAuth as the authentication and Streamable HTTP as the transport. Leave Client ID and Client Secret empty, as the server provides the sign-in itself.
  4. Confirm that you have read about the risks of custom connectors, and click Add.
  5. Click the connector and sign in with your Placepoint user.
  6. Repeat for the other servers.

On Enterprise, an administrator can add the servers for the whole organisation. Each user still signs in with their own Placepoint user, see Which account do I sign in with?.

Google Gemini Enterprise

Placepoint AI can be used from Gemini Enterprise in Google Cloud. An administrator adds each server as a data store of the type Custom MCP Server, see Set up a custom MCP server. Google labels the feature as a preview.

Before you start, the administrator needs the Discovery Engine Editor role. The organisation policy in Google Cloud blocks custom MCP servers by default, so it must be overridden for the domain placepoint.no.

  1. Open Gemini Enterprise in the Google Cloud console, select Data stores and click Create data store.
  2. Search for Custom MCP Server and click Add MCP server.
  3. Select OAuth 2.0 under Authentication settings and fill in the fields:
    • MCP Server URL: https://mcp.placepoint.no for Placepoint MCP, https://mcp-dataset.placepoint.no for Placepoint Dataset MCP.
      • Authorization URL: the same URL with /authorize at the end.
      • Token URL: the same URL with /token at the end.
      • Client ID and Client Secret: type placepoint in both. The server uses its own client and ignores what is entered here.
      • Tick Enable PKCE Support.
      • Scopes: openid offline_access.
  4. Click Verify Auth, sign in with your Placepoint user and click Continue twice.
  5. Choose a region, give the data store a name and click Create. Wait until the status is Active.
  6. Open the data store, select Actions and Reload custom actions. All tools are switched off to begin with: select the ones you want to use and click Enable actions. Gemini Enterprise allows up to 100 active tools per server.
  7. Repeat for Placepoint Dataset MCP.

Then connect the data store to the app in Gemini Enterprise. Each user signs in with their own Placepoint user the first time, see Which account do I sign in with?.

Google Gemini CLI

Gemini CLI connects from the terminal. Add the servers with one command per server:

gemini mcp add --scope user --transport http placepoint https://mcp.placepoint.no
gemini mcp add --scope user --transport http placepoint-dataset https://mcp-dataset.placepoint.no

Start gemini, type /mcp auth placepoint and sign in in the browser that opens. Repeat with /mcp auth placepoint-dataset. With --scope user the servers apply in every folder. Without it they end up in the folder you are in, and Gemini CLI switches off MCP servers in folders you have not marked as trusted.

The server name cannot contain an underscore, because Gemini CLI uses it in the name of every tool.

Other MCP clients

The servers use Streamable HTTP with OAuth sign-in that the client discovers itself. Other clients that support that should therefore also work, but we have not tested them. The Gemini app at gemini.google.com can connect to custom MCP servers only from personal Google accounts in the USA, so we have not tested it.

Take questions, missing access and feedback up with your contact at Placepoint. If you do not have a contact yet, get in touch via placepoint.no/selskap/kontakt.

The tools

Placepoint MCP has around 100 tools and Placepoint Dataset MCP around 70. In Claude the agent chooses among all of them itself, but in Microsoft Copilot Studio and other clients with a limit on the number of tools, you must choose which ones the agent should have. The tools in Placepoint AI shows what each tool does, with an example of a question it answers.

The tools are grouped by task:

  • Placepoint MCP: property search and lookups, addresses and places, companies and ownership, land register and changes of ownership, land-use plans and zoning, map layers and risk, maps and traffic, travel time, reports and geometry.
  • Placepoint Dataset MCP: projects and datasets, file upload, linking to properties, map style, icons, administrative areas, travel time and geometry.

Always leave get_usage_guide and get_tool_help switched on in both servers, and get_mcp_context in Placepoint Dataset MCP. They tell the agent how to use the other tools.

Caveats

  • Placepoint AI is under active development. Some tools and map layers are not fully tested and can fail.
  • All the servers require sign-in with your Placepoint user. Access follows the user, so it cannot be lent out by sharing a URL.
  • The language model can be wrong. The results are a starting point, not a final answer, and must be verified against official sources before they are used for decisions.
  • Access, features and data can change along the way.
  • Claude, Copilot Studio, ChatGPT, Codex, Cursor, VS Code, Perplexity, Gemini Enterprise and Gemini CLI are other vendors' products. We help with what we develop and deliver ourselves, Placepoint MCP, Placepoint Dataset MCP and Combify MCP, not with licences, setup or faults in those products. There we refer you to the vendor's own documentation and support.