Your CRM now understands what you ask.
Topyn's AI assistant looks things up, summarizes, drafts to-dos and updates records – with your own permissions. The remote MCP server puts the same power in Claude or any MCP client.
- Works with your own permissions
- Risky steps need your approval
- Every tool call is logged
It works with the same permissions you have – through the same endpoints, into the same change log. And at the high-risk steps listed below, it stops and waits for your approval.
01 /Surfaces
Two doors, one rulebook.
One toolkit serves two surfaces: the built-in assistant inside the app, and the remote MCP server from Claude.
Built-in assistant – in the app, one click away
- Sees what you see: the workspace snapshot and the record you have open
- Shows its work, step by step
- Conversations stick around – history, renaming, deleting
- ⌘J / Ctrl+J from anywhere
Remote MCP – from Claude too
- claude.ai, Claude Desktop, Claude Code, or any MCP client
- OAuth 2.1 with your own account – read, and write if you allow it
- Pro and Enterprise plans; off by default, the admin turns it on
api.topyn.hu/mcp
The whole CRM, not just deals.
One shared toolkit serves the built-in assistant and the MCP server. Every module is reached through its own endpoints – whatever you have no access to, the AI cannot see either.
30+ modules
modules
700+ operations
operations
1 rulebook
rulebook
Depends on the modules in your plan and your own permissions.
02 /Examples
What can you ask?
What's available depends on your plan, your enabled modules and your permissions.
03 /How it works
How it works in practice.
A request's path from Claude to your pipeline – and back.
The client registers, and you decide on Topyn's own approval screen: read, and write if you want.
The client only receives the tools your role and your enabled modules allow.
The same endpoint, permission check, validation and audit trail run as when you click in the UI.
The result comes back with links – one click, and you are at the record.
For a risky step, nothing runs: the client gets a link, and you decide in Topyn, within 24 hours.
The client registers, and you decide on Topyn's own approval screen: read, and write if you want.
04 /Connect
Connect in a few steps.
- 1
Pro or Enterprise plan
- 2
The admin turns it on: Settings › AI › MCP connection
- 3
Everyone connects with their own account
- 1In Claude's settings, under Connectors, add a custom connector: name it "Topyn", and use the address https://api.topyn.hu/mcp.
- 2After connecting, Topyn's own approval screen opens — you decide what to allow.
- 3Click Allow, and Claude can reach Topyn's tools in the conversation.
Claude Connectors — Claude's own menus depend on Anthropic's interface and may change.
05 /Safety
You decide. The AI can only do what you can.
With your own permissions
The AI works through the same endpoints as the UI: role, module access, validation and audit trail all apply.
Only your own workspace
Every connection belongs to one user and one workspace; the approval screen shows exactly which.
A risky step needs confirmation
Only the irreversible categories listed below ask for confirmation – then the client gets a link, and you decide.
Every other write runs immediately – exactly as if you had clicked it yourself.
Read or write – your call
Reading is always granted; writing only if you check the box at approval. Without write access, the client does not even see the write tools.
Off by default, stoppable anytime
The admin can switch MCP off with one toggle – connections pause until then. Revoking is immediate and final.
Every call logged
Tool calls are logged, and changes land in the audit trail – exactly as from the UI.
The workspace's Settings › AI page shows the AI processing's data recipients and the consent given.
06 /FAQ
Frequently asked questions about the AI assistant and MCP
Which plan includes MCP?
The Pro and Enterprise plans. It is off by default; the workspace admin turns it on under Settings › AI.
Does everyone have to connect separately?
Yes. Everyone approves the connection with their own Topyn account, so the AI works with exactly their permissions.
What can Claude see of my data?
Only what a tool call actually queries, and only what you yourself have access to. Client-side data handling is governed by that client's own provider's terms.
What happens if the AI wants to do something risky?
The action does not run. The built-in assistant shows a confirmation card; on MCP the client gets an approval link — the request expires after 24 hours.
How do I disconnect?
Settings › MCP › Revoke. An admin can revoke anyone's connection.
Does it only work with Claude?
No. Any MCP client that supports Streamable HTTP and OAuth 2.1 with dynamic client registration can connect.
Do I need a Claude subscription?
Custom connector availability is Anthropic's call – check Claude's own help center for which plan includes it.
Ask. Connect. Approve.
Free for 14 days, no card required.