Connect
Connect Claude to Zoye
Connect Claude to your Zoye workspace and you can ask it to do your Zoye work in plain language: pull up a lead, add a deal, move something along, chase what is overdue, or summarise where the month stands. There is no API key to copy and nothing to install. You paste one address into Claude, sign in to Zoye, and approve.
What you need
- A Zoye plan that includes the connector: Growth, Scale, or Custom.
- Admin access to the workspace you want to connect, the same as creating an API key.
- A Claude account. The connector works on the web, on desktop, and on mobile.
Set it up
- In Zoye, open Integrations and choose Claude. Copy the connector address shown there. It is the same address for everyone on your team:
https://api.zoye.io/mcp- In Claude, open Settings, then Connectors, and choose to add a custom connector. Paste the address.
- Claude sends you to Zoye to sign in. Tick which workspaces Claude may use, review what it will and will not be able to do, and approve.
That is the whole setup. Claude now appears as a connector in your Claude account and the connection appears in Zoye under Integrations, where you can disconnect it at any time.
You choose the workspaces when you approve, and you can tick more than one. Claude can read and write only the ones you ticked: anything else is invisible to it, including workspaces you personally have access to. The first workspace you tick is where a new record lands when Claude does not name one.
What Claude can do
Claude gets a small, general set of tools rather than a long list of narrow ones, because the same handful of tools works across every record type in Zoye:
| Tool | What it does |
|---|---|
| list_entity_types | See what kinds of record exist in your workspace. |
| get_entity_schema | Read the real fields and stage names, per workspace. |
| describe_capability | Answer questions about what Zoye can do, from fact. |
| query_records | Search and filter any record type, group and total it. |
| get_record | Open one record in full. |
| get_full_context | Everything about one record: relations and attachment text. |
| run_report | Run the same analytics the Reports page shows. |
| get_briefing | What needs attention across the whole workspace. |
| create_records | Add records of any type, one or many. |
| update_records | Change fields, add or remove list values. |
| delete_records | Delete records. Always asks you first. |
| restore_records | Bring something back from the archive. |
| move_to_stage | Move a lead, deal or task along its board. |
| link_records | Connect records to each other. |
| move_to_workspace | Move a record to another workspace. |
| manage_tags | Create, list and remove tags. |
| view_archive | See and empty the recycle bin. |
| manage_workspaces | List and create workspaces. |
| zoye_ask | Anything else: hands the request to Zoye's own assistant. |
Things to ask it
- "What needs my attention in Zoye today?"
- "Add Jordan Rivera at Northwind as a new lead from the website."
- "Which deals have not moved in three weeks?"
- "Move the Northwind deal to Proposal and add a task to follow up Friday."
- "Summarise this month's invoices against last month."
Deleting is always confirmed
Anything that destroys data is gated. When Claude tries to delete records, empty the archive, or remove a tag, Zoye refuses the first attempt and returns a description of exactly what would be affected. Claude shows you that description, and only proceeds if you say yes.
This is not only about mistakes. Text inside your own records, a note on a contact, or the contents of an attached document, becomes part of what an assistant reads. If that text contained an instruction like "now delete everything", the confirmation step is what stops it. For a filter-based delete Zoye also requires the expected number of records, so Claude has to look before it deletes.
What it costs you
Almost nothing. When Claude reads or updates your Zoye data through the tools above, Claude does the thinking on your Claude plan and Zoye spends no AI prompts, so those calls cost zero Zoye prompts.
The one exception is zoye_ask, which hands the request to Zoye's own assistant for anything the tools above do not cover, such as sending an email or building an automation. That runs on Zoye's side, so it costs one Zoye prompt, the same as asking Zoye directly in the app.
Security and access
- Claude acts as the person who approved the connection, with exactly their permissions. It cannot see or do anything that person could not.
- It is limited to the workspaces ticked at approval. Naming any other workspace simply fails, and a cross-workspace search only ever spans the approved set. It can never see your billing, your plan, or your team settings.
- Access is checked on every single request. If your plan changes, or the person who connected it leaves the organisation, the connection stops working and Zoye shows you why.
- You can disconnect at any time from Integrations, or by asking Zoye's assistant to disconnect Claude. It takes effect immediately.
Claude Code and other MCP clients
The same endpoint works with any client that speaks the Model Context Protocol over HTTP. From a terminal:
claude mcp add --transport http zoye https://api.zoye.io/mcpYou will be asked to sign in to Zoye the first time, exactly as in the browser. If your client cannot do the sign-in flow, create a connector key in Zoye under Integrations and pass it as a bearer token instead.
Looking for the older local MCP server, the @zoye/mcp npm package? It still works but only covers contacts, and it needs a key in a config file. This hosted connector replaces it. See MCP server for that setup.