Kelpic

Kelpic® Documentation

Connect Kelpic® to Your AI Assistant

Ask your own AI about your properties, and it answers from your real data instead of guessing. Setup takes about two minutes, there is no key to generate, and the steps for every assistant are on this page.

Kelpic®'s address for your assistant

https://kelpic.com/api/v1/mcp

Whatever your assistant calls it — connector, tool, integration or MCP server — this is the only address you need.

Pick your assistant

Each one has its own steps below. Not listed? Use any other client — Kelpic follows the open standard, so anything that speaks it will work.

Claude Claude Code ChatGPT Google Gemini CLI Microsoft Copilot Studio Cursor VS Code Any other client

Before you start

  • A Kelpic account you can sign in to, on the organization whose properties you want to ask about.
  • A desktop browser on the machine you are setting this up on. The sign-in opens a browser window and comes back to it.
  • Nothing else. No API key to generate, no client secret to copy, no developer account to sign up for.

Step-by-step setup

Claude

Claude on the web, Claude Desktop and Cowork all use the same Connectors screen. Custom connectors work on Free, Pro, Max, Team and Enterprise, though the Free plan allows only one.

  1. 1

    Open your connector settings

    In Claude, open Settings and go to Connectors. If you are on a Team or Enterprise plan and you own the organization, use Organization settings → Connectors instead, so everyone gets it at once.

  2. 2

    Add a custom connector

    Click the + button and choose Add custom connector. Organization owners click Add, hover over Custom, and pick Web.

  3. 3

    Paste Kelpic's address

    Enter the address below as the remote MCP server URL. Leave Advanced settings empty — Kelpic registers your assistant automatically, so there is no client ID or secret to fill in. 

    https://kelpic.com/api/v1/mcp
  4. 4

    Click Add, then sign in

    A Kelpic window opens. Sign in if you are not already, choose whether the assistant may only read or also make changes, and press Approve.

  5. 5

    Turn it on in a conversation

    Claude enables connectors per conversation. In the chat box, click +, choose Connectors, and switch Kelpic on. Ask it something from the list further down this page to confirm it works.

Claude Code

One command in your terminal, then a browser sign-in. Add --scope user to make Kelpic available in every project instead of only the current one.

  1. 1

    Register the server

    Run this in any terminal: 

    claude mcp add --transport http kelpic https://kelpic.com/api/v1/mcp
  2. 2

    Authenticate

    Start Claude Code, run /mcp, select kelpic and choose Authenticate. Your browser opens Kelpic's sign-in and consent screen. 

    /mcp
  3. 3

    Check it connected

    Run /mcp again. Kelpic should be listed as connected, with its tools available. On a machine with no browser, authenticate on one that has a browser — the sign-in has to open a real window.

ChatGPT

ChatGPT reaches custom MCP servers through developer mode, in the web app. It is available on Plus, Pro, Business, Enterprise and Edu plans.

  1. 1

    Turn on developer mode

    Open Settings → Apps & Connectors → Advanced settings and switch on Developer mode. On a Business or Enterprise workspace an administrator has to allow it first, under Workspace settings → Permissions & roles.

  2. 2

    Create the connector

    Go back to Apps & Connectors and click Create.

  3. 3

    Fill in the details

    Name it Kelpic, paste the address below as the MCP server URL, and choose OAuth as the authentication method. There is no key to paste anywhere. 

    https://kelpic.com/api/v1/mcp
  4. 4

    Sign in and approve

    Click Create. Kelpic's sign-in opens; approve the access you want to give. ChatGPT warns that custom connectors are not verified by OpenAI — that warning appears for every custom server, Kelpic included.

  5. 5

    Use it in a chat

    Open the apps or + menu inside a conversation and select Kelpic before you ask your question.

Google Gemini CLI

Gemini CLI reads its MCP servers from a settings file. Use ~/.gemini/settings.json for every project, or .gemini/settings.json inside one repository.

  1. 1

    Open your settings file

    Open ~/.gemini/settings.json in an editor, creating it if it does not exist yet.

  2. 2

    Add Kelpic

    Add an mcpServers entry. If the file already has one, add kelpic alongside your existing servers rather than replacing the block. 

    {
  "mcpServers": {
    "kelpic": {
      "httpUrl": "https://kelpic.com/api/v1/mcp"
    }
  }
}
  3. 3

    Authenticate

    Save the file and start gemini. Run /mcp to list your servers; if Kelpic is not signed in yet, run the command below and approve in the browser window that opens. 

    /mcp auth kelpic
  4. 4

    A note on servers and containers

    The sign-in redirects back to a local address, so it needs a desktop browser on the same machine. On a headless box, set the connection up locally first.

Microsoft Copilot Studio

Copilot Studio treats Kelpic as a tool you attach to one of your agents.

  1. 1

    Open your agent's tools

    In Copilot Studio, open the agent you want to give access to, go to its Tools page and choose Add a tool.

  2. 2

    Choose Model Context Protocol

    Select Model Context Protocol. Kelpic is not in the built-in connector list, so choose the option to create a new MCP server.

  3. 3

    Describe the server

    Name it Kelpic, give it a short description such as “Property, lease, tenant and maintenance data”, and paste the address below as the server URL. 

    https://kelpic.com/api/v1/mcp
  4. 4

    Set authentication to OAuth 2.0

    Choose OAuth 2.0 with dynamic discovery. Kelpic supports dynamic client registration, so Copilot Studio sets itself up and you never create a client ID or secret.

  5. 5

    Add it to the agent and connect

    Create the tool, then choose Add to agent. The first time the agent uses it you are asked to connect — sign in to Kelpic and approve.

Cursor

Cursor keeps MCP servers in an mcp.json file: ~/.cursor/mcp.json applies everywhere, .cursor/mcp.json applies to one project.

  1. 1

    Open the MCP settings

    Press Ctrl+, (Cmd+, on a Mac), go to Tools & MCP in the sidebar and click New MCP Server. Cursor opens mcp.json for you. Creating the file yourself works just as well.

  2. 2

    Add Kelpic

    Add the entry below, keeping any servers you already have. 

    {
  "mcpServers": {
    "kelpic": {
      "url": "https://kelpic.com/api/v1/mcp"
    }
  }
}
  3. 3

    Sign in

    Save the file and return to Tools & MCP. Kelpic appears in the list; enable it and click the login prompt. Kelpic's sign-in opens in your browser.

  4. 4

    Confirm the tools loaded

    The Tools & MCP screen lists Kelpic's tools once the connection succeeds. If the list is empty, the sign-in did not finish — click the login prompt again.

VS Code

VS Code uses Kelpic in Copilot's agent mode. There is a command for this, so you rarely need to edit the file by hand.

  1. 1

    Run the add-server command

    Press Ctrl+Shift+P (Cmd+Shift+P on a Mac), run MCP: Add Server, and choose HTTP.

  2. 2

    Paste the address and name it

    Enter the address below, call the server kelpic, and choose whether to save it for this workspace or for you globally. 

    https://kelpic.com/api/v1/mcp
  3. 3

    Or edit mcp.json directly

    The command writes .vscode/mcp.json for a workspace, or your user profile's copy. Editing it by hand gives the same result: 

    {
  "servers": {
    "kelpic": {
      "type": "http",
      "url": "https://kelpic.com/api/v1/mcp"
    }
  }
}
  4. 4

    Allow the sign-in

    Start the server. VS Code says the MCP server definition wants to authenticate — choose Allow, then sign in to Kelpic and approve.

  5. 5

    Pick Kelpic in chat

    Open Copilot Chat in agent mode and use the tools picker to make sure Kelpic's tools are selected.

Any other client

Kelpic is a standard remote MCP server. If you are wiring up a client that is not listed here, or building your own agent, this is everything you need.

  1. 1

    Endpoint and transport

    Streamable HTTP, at the address below. Point your client's remote or HTTP server setting at it. 

    https://kelpic.com/api/v1/mcp
  2. 2

    Authorization

    OAuth 2.1, authorization code with PKCE (S256). Kelpic supports dynamic client registration, so a client that follows the spec needs no key, no secret and no manual registration from you.

  3. 3

    Discovery documents

    A compliant client finds everything it needs from these two, starting with the 401 challenge the endpoint returns: 

    https://kelpic.com/.well-known/oauth-protected-resource/api/v1/mcp
https://kelpic.com/.well-known/oauth-authorization-server
  4. 4

    Scopes

    kelpic:read for reading, kelpic:write for changes. Ask only for what you need — a connection granted read access is never offered the tools that change anything. 

    kelpic:read kelpic:write
  5. 5

    Tokens

    Access tokens are bound to this endpoint as their audience, so a token issued for Kelpic's MCP server cannot be replayed against the REST API, or the other way round. They expire after an hour and refresh automatically.

What happens when you approve

Whichever assistant you used, the last step is the same. Kelpic shows you which assistant is asking and exactly what it wants, and nothing is granted until you press Approve.

Read only

The assistant can look things up and answer questions. The tools that change anything are not offered to it at all, so there is nothing for it to get wrong.

Read and change

It can also update a unit, adjust rent on a lease or open a maintenance request. Your assistant shows you each change and waits for your approval first.

Start with read only. You can disconnect and connect again with more access whenever you decide you want it.

What you can ask it

Once connected, your assistant can look things up for itself. You ask in ordinary language; it reads the same records you would have opened.

  • “Who is behind on rent, and by how much?”

    Reads your bills, due dates and payment status.

  • “Summarize every open maintenance request across my properties.”

    Reads requests with their unit, status and assignee.

  • “Which leases end in the next 90 days?”

    Reads lease terms, rent and renewal settings.

  • “Compare the applications on my Maple Court duplex.”

    Reads applicant details, income and screening consent status.

  • “Raise the rent on unit 4B to $1,450.”

    A change like this needs write access, and your assistant asks you to confirm first.

  • “Open a maintenance request for the leaking tap in unit 2A.”

    Filed under your name, the same as if you had created it yourself.

What it can reach — and what it cannot

A connected assistant sees your organization and nothing else. It inherits your own permissions: it can never reach further than you can.

It can read

  • Properties, buildings, units and floor plans
  • Leases, rent, deposits and late-fee settings
  • Tenants and rental applications
  • Maintenance requests
  • Rent bills, due dates and how far behind they are

It can never

  • See any organization other than yours
  • Change where your money goes, or your payout and fee settings
  • Change a lease's dates, its unit, or who is on it
  • Delete anything
  • Do more than the person who connected it can already do

Staying in control

Reading is the default
Permission to change anything is separate, and you grant it deliberately. A read-only connection is not simply refused when it tries to make a change — the tools that make changes are not offered to it at all.
Disconnecting takes effect immediately
Remove the connector in your assistant — Connectors in Claude, Apps & Connectors in ChatGPT, the kelpic entry in your mcp.json elsewhere. Kelpic re-checks on every single request that the connection is still active, so access ends at once rather than whenever a session happens to expire.
Connections belong to one person
Connecting your own assistant does not give your colleagues' AI access to anything, and their connections are theirs to manage.
Your assistant asks before it acts
Where a change is requested, AI assistants show you what they are about to do and wait for your approval. That prompt comes from your assistant, so exactly how it looks depends on which one you use.

If something does not work

My assistant says it cannot find the server

Almost always a typo in the address. It has to be exactly https://kelpic.com/api/v1/mcp — https, no trailing slash, nothing added on the end. Use the copy button rather than typing it out.

I get an error mentioning “audience” or an invalid token

Your client is dialling a different host than the one Kelpic issues tokens for. Remove the connector, add it again using the exact address above, and sign in once more.

The sign-in window opens and closes with nothing happening

Check your pop-up blocker, and make sure you are signed in to the Kelpic account that owns the properties you want to ask about. If it is a different account, sign out of Kelpic in that browser and start the connection again.

It connected, but it says it has no tools

Some assistants enable a connector per conversation rather than once and for all. In Claude, use the + button in the chat and switch Kelpic on. In VS Code, select Kelpic in the agent-mode tools picker.

It refuses to change anything

The connection was approved for reading only. That is deliberate — the tools that make changes are never offered to a read-only connection. Disconnect it in your assistant, connect again, and approve reading and changing.

I am setting this up on a server or in a container

Signing in needs a real browser on the same machine, because the approval redirects back to a local address. Set the connection up on your own computer instead.

The menus in my assistant do not match these steps

Every AI company redesigns its settings from time to time, and we cannot always update the same week they do. Look for anything called connectors, tools, integrations or MCP servers. The address is the only part that has to be exact — and if you find the steps have moved, tell us and we will fix this page.

Need a hand?

Tell us which assistant you are using and what it said, and we will walk you through it.

Contact support