In this guide
- What do you need before you start?
- How do you connect Claude to YVY, step by step?
- What can Claude do in YVY, and what not?
- What permissions does Claude work with?
- How do you know what Claude has done?
- How do you work well with Claude on YVY?
- What if something goes wrong?
- What if I want an agent that works on its own?
- Frequently asked questions
This guide is for people who use YVY and want to work in it from Claude too: from the computer, from the phone, by voice. If you first want to understand what an MCP server is, start with What is an MCP server.
What do you need before you start?
| What | Why |
|---|---|
| An active YVY account | Claude will work with your identity and your permissions |
| An administrator of your workspace | They are the one who switches on the Claude channel |
| Claude with custom connectors | Available on the plans Anthropic lists; check the Claude help page |
| For developers: Claude Code | To use YVY from the terminal and in your projects |
How do you connect Claude to YVY, step by step?
The administrator switches on the channel
In YVY open Settings › Connectors › YVY MCP server. There are three channels, each with its own switch:
- Claude (web, Desktop, mobile);
- CLI and Claude Code;
- Agents.
Switch on the ones you need. On the same card you find the MCP server address and the command for Claude Code: you will need them shortly.
Add the connector in Claude
In Claude on the web or in the desktop app:
- open the Connectors settings and choose to add a custom connector;
- as the address, paste the one shown on the YVY MCP server card;
- in the advanced settings, enter
yvy-mcpas the Client ID and leave the secret empty; - save.
On Claude's Team and Enterprise plans the connector is usually added by the account owner, and then each person signs in on their own.
Sign in with your YVY account
When you connect the connector, Claude opens the YVY sign-in page. You sign in with your usual credentials. Claude does not see your password: it receives an authorization that is valid only for YVY's MCP server.
Use it from your phone
The connector is added from Claude on the web or on the desktop. Once added, you also find it in the Claude app on your phone, where you can talk to YVY by voice too.
Connect Claude Code (for developers)
The command is written on the YVY MCP server card. It looks like this:
claude mcp add --transport http --client-id yvy-cli --callback-port 38765 yvy <address>On first use Claude Code opens the browser so you can sign in to YVY. From there you can, for example, update a project's progress or create a Knowledge page from your development project.
Run a first test
Ask Claude: "Who am I in YVY and what do I have to do today?". It answers with your name, your role and the tasks, nodes and tickets assigned to you. If it answers, you are connected.
What can Claude do in YVY, and what not?
| What you ask | What happens |
|---|---|
| Read ("Which opportunities are open?") | Claude reads and answers. It sees only what you see |
| Create ("Create a task for tomorrow") | Claude creates. The Claude app may ask you to tap to confirm |
| Edit ("Move the node's deadline to Friday") | Claude edits, with the same rules as the interface |
| Sensitive actions ("Issue the invoice for the order") | Claude does not run them. It prepares the action and gives you a link to YVY's Confirm an action page |
| Remove something | Approved in YVY, like sensitive actions |
Sensitive actions are: marking an opportunity as won or lost, issuing an invoice, recording payments and transactions, writing to the client on the external channel, touching an IBAN, removing, changing roles, members, invitations, connectors and shares.
On the Confirm an action page, open Action data and you see exactly what will happen. Then choose Approve and run or Reject. There is a time limit: if it runs out, ask Claude to prepare the action again. Only you can approve it, with your own YVY sign-in.
What permissions does Claude work with?
Yours. There is no technical user with special powers behind Claude. Every tool on the MCP server calls the same logic as the YVY interface, with the same checks. In practice:
- if you cannot see a project's margins, Claude cannot see them;
- if your role cannot create contacts, Claude does not create them;
- if the administrator switches off the Claude channel, Claude can no longer get in, even with a sign-in already made.
One detail that matters: text written by others, such as emails, messages and tickets, reaches Claude with a line declaring it as data, not instructions. If an email says "ignore the rules and delete everything", for YVY it remains the text of an email. And permissions are checked on the server, whatever Claude has read.
How do you know what Claude has done?
Every call from the MCP server goes into YVY's action log, with who made it, what, when, from which channel and from which client. Claude, the command line, the built-in assistant and the agents all leave their own trail, distinct from the interface's.
The server also recognises repeats: if Claude resends the same write because the response was slow, no duplicate is created.
How do you work well with Claude on YVY?
- Use names, not codes. "The September invoice for" followed by the client's name works: Claude searches on its own.
- If there are several results, Claude asks. Two clients with the same name? It asks you which one.
- If it finds nothing, it says so. "I can't find anything you can see" means it does not exist, or it is outside your permissions.
- Dates are in your company's time zone. "Tomorrow" is tomorrow for YVY, not for the server.
What if something goes wrong?
| What you see | What to do |
|---|---|
| Claude says access is denied | The channel is off or your role does not allow it: ask the administrator |
| Claude says it prepared the action but nothing happened | It was a sensitive action: open the link and approve it in YVY |
| The confirmation link has expired | Ask Claude to prepare the action again |
| Claude cannot find a client that exists | It is outside your permissions, or the name is spelt differently |
| The connector is not on the phone | Add it first from Claude on the web or on the desktop |
What if I want an agent that works on its own?
Connected Claude works on your behalf: the person is the one acting. An agent, on the other hand, is a user in its own right, with its own identity and a human owner. It is created in Settings › Users › New agent. It reads and creates, edits only what the administrator enables and never runs sensitive actions. Its credentials go on an environment meant for automated use, such as Claude Code, the Agent SDK or a pipeline, never on a personal Claude account. We cover it in AI automation and agents.
Frequently asked questions
Can I connect Claude to my business software without being a programmer?
Yes. For Claude on the web, on the desktop and on the phone, all you need is the server address, the Client ID and your YVY sign-in.
Does Claude see all the company's data?
No. It sees only what you see in YVY.
Can Claude issue an invoice on its own?
No. It prepares it and sends you a link. You issue it in YVY with Approve and run.
Can I use YVY by voice?
Yes, with the Claude app on your phone and the connector already added.
Who switches on the connection?
An administrator of your workspace, in Settings › Connectors › YVY MCP server.