MCP interface and API for AI tools

Supported AI tools

One connection, all common AI tools.

MCP, the Model Context Protocol, is the open standard for this. Anything that speaks it can be connected to HerzensManagement.

Claude Desktop Anthropic As a connector in the settings
Claude Code Anthropic One command in the terminal
ChatGPT OpenAI As a custom connector
Cursor Anysphere Entry in mcp.json
GitHub Copilot in VS Code MCP server in agent mode
Gemini CLI Google Entry in settings.json
Windsurf Cognition MCP server in Cascade
Le Chat Mistral AI As a custom connector
n8n Automation MCP client node in the workflow
and any other tool that supports MCP

How it works

Ask instead of clicking.

The AI tool decides for itself which tools it needs for your question, calls them at HerzensManagement and answers with your agency’s data.

You ask

In plain language, in the tool you already use.

The tool calls tools

For example list_shifts or list_invoices. All tools are read-only.

HerzensManagement delivers

Only your agency’s data, and only from the areas the key is permitted to read.

Use cases

What agencies and teams can use it for.

Weekly overview without clicking

“Which rotations change next week?”

The AI tool reads the rotations in the period and lists which clients have a change coming up and which caregiver is assigned.

Unpaid invoices at a glance

“Which invoices are still unpaid?”

Query invoices with status, amount and period and have the result summarised.

Clarify availability

“Which caregivers are available from 1 November?”

The AI tool reads caregivers and their rotations and names who is not assigned from that date.

Monthly overview for management

“Create the overview for October for me.”

Clients by status, rotations and invoices for the month as a report, in the format your AI tool handles best.

Your own workflows in n8n

Send the week’s rotation changes to the team every Monday.

In n8n the same tools serve as building blocks of a workflow. HerzensManagement supplies the data; what happens next is up to your workflow.

For your developers

“Show me a client’s rotations in December.”

Anyone connecting their own application has the agency’s data to hand directly in Claude Code or Cursor.

Setup

Create a key, enter it, connected.

Once the interface has been activated, your agency admin creates a key in the panel under “Einstellungen → API” (Settings → API): with a name, a selection of permissions and an optional expiry date. The key is shown exactly once and can be revoked at any time.

The address and key in the examples are placeholders. The real address of your interface is shown in the panel under “Einstellungen → API”.

Claude Desktop

claude_desktop_config.json

{
  "mcpServers": {
    "herzensmanagement": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://IHRE-API-ADRESSE/api/mcp",
        "--header", "Authorization: Bearer ${HERZENS_API_KEY}"
      ],
      "env": { "HERZENS_API_KEY": "IHR-SCHLUESSEL" }
    }
  }
}

Settings → Developer → Edit config. Then restart Claude.

Claude Code

Terminal

claude mcp add --transport http herzensmanagement \
  https://IHRE-API-ADRESSE/api/mcp \
  --header "Authorization: Bearer IHR-SCHLUESSEL"

One command in the project folder. Afterwards, check the connection with /mcp.

ChatGPT

Settings → Connectors

Name:            HerzensManagement
MCP server URL:  https://IHRE-API-ADRESSE/api/mcp
Authentication:  API key (IHR-SCHLUESSEL)

Then select the “HerzensManagement” connector in the chat.

Custom connectors require developer mode in your ChatGPT workspace.

Cursor

.cursor/mcp.json

{
  "mcpServers": {
    "herzensmanagement": {
      "url": "https://IHRE-API-ADRESSE/api/mcp",
      "headers": { "Authorization": "Bearer IHR-SCHLUESSEL" }
    }
  }
}

Applies to the project; under ~/.cursor/mcp.json for all projects.

Direct (JSON-RPC)

Terminal

curl -X POST https://IHRE-API-ADRESSE/api/mcp \
  -H "Authorization: Bearer IHR-SCHLUESSEL" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/list"
  }'

MCP is an open standard. This is how you query, without an AI tool, which tools your key may use.

Transport: MCP over HTTP (streamable HTTP), authentication with Authorization: Bearer <key>.

The same data is also available as a plain REST API (/api/v1/..., Bearer key). The developer documentation with a quick start and examples is in the panel next to the key settings.

Tools

What your AI tool can read in HerzensManagement.

Seven tools, all strictly read-only. There are no writing tools. You define which areas a key may read through its permissions:

clients:read caregivers:read shifts:read invoices:read

Tool	What it returns	Access
`get_agency`	Master data of your agency	Read
`list_clients`	Clients with status, place and care level; filter by status, search	Read
`get_client`	A single client	Read
`list_caregivers`	Caregivers; filter by status, search	Read
`get_caregiver`	A single caregiver	Read
`list_shifts`	Rotations in a date range, per client or caregiver	Read
`list_invoices`	Invoices with status, amount and period	Read

You decide what an AI tool sees.

Each key applies only to your one agency, is read-only and can be revoked at any time. Whatever a tool returns is processed by the provider of your AI tool. For personal data of clients and caregivers, including health-related information, you need a data processing agreement with that provider. Through the permissions you can restrict a key to less sensitive areas.

Book a demo

Frequently asked questions

What is MCP?

The Model Context Protocol is an open standard through which AI tools such as Claude or ChatGPT access external systems. A service describes its tools there, and the AI tool calls them when needed. You do not need any programming of your own for this.

Which AI tool do I need?

Any tool that supports MCP. These include Claude Desktop, Claude Code, ChatGPT, Cursor, GitHub Copilot in VS Code, Gemini CLI, Windsurf, Le Chat and n8n. Any other MCP-compatible tool can be connected as well.

Can the AI tool change anything?

No. All tools are read-only. There are no writing tools.

Which data can the AI tool access?

The agency’s master data, clients, caregivers, rotations and invoices, and only those of the agency the key belongs to. With the permissions clients:read, caregivers:read, shifts:read and invoices:read you define for each key which areas it may read.

What does this mean for data protection?

Whatever a tool returns is processed by the provider of your AI tool. For personal data of clients and caregivers, including health-related information, you therefore need a data processing agreement with that provider. Through the permissions you can restrict a key to less sensitive areas.

How do I get access?

The interface is switched off by default. On request, the HerzensManagement team activates it for your agency. After that, your agency admin creates a key in the panel under “Einstellungen → API” (Settings → API): with a name, permissions and an optional expiry date. The key is shown exactly once and can be revoked at any time.

Is there a plain API as well?

Yes. The same data is available as a REST API under /api/v1/..., authenticated with a Bearer key. The developer documentation with a quick start and examples is in the panel next to the key settings.

Connect HerzensManagement to any AI tool