Connect the Service4AI MCP server to your agent

At a glance

Claude Code

One command in your terminal:

claude mcp add --transport http service4ai https://mcp.service4ai.com/v1/en

Add --scope user to make it available in all your projects. In a session, the /mcp command shows the server status and its five tools.

Claude (web and desktop app)

Remote MCP servers are added as custom connectors, depending on your plan:

1. Open Settings → Connectors.
2. Choose Add custom connector.
3. Name: Service4AI. URL: https://mcp.service4ai.com/v1/en. Leave the OAuth settings empty.
4. In a conversation, turn the connector on from the tools menu.

The connector is then available on claude.ai and in the desktop app.

Cursor

Add the server to ~/.cursor/mcp.json (all projects) or to .cursor/mcp.json at the root of a project:

{
  "mcpServers": {
    "service4ai": {
      "url": "https://mcp.service4ai.com/v1/en"
    }
  }
}

The server then shows up in Cursor's MCP settings, where you can turn it on or off.

VS Code (GitHub Copilot)

Create .vscode/mcp.json in your project:

{
  "servers": {
    "service4ai": {
      "type": "http",
      "url": "https://mcp.service4ai.com/v1/en"
    }
  }
}

The tools are then offered to Copilot Chat in agent mode.

ChatGPT

ChatGPT accepts remote MCP servers in developer mode, depending on your plan:

1. Turn on developer mode in the connectors (or apps) settings.
2. Create a connector with the URL https://mcp.service4ai.com/v1/en and authentication set to None.
3. Turn it on in a conversation. ChatGPT may ask for confirmation before tools that change something (renting or cancelling).

ChatGPT's menus change often: if a label differs, look for “MCP” or “connector” in the settings.

Other clients

Any client that supports the Streamable HTTP transport works with the URL alone. For a client that only knows local (stdio) servers, go through the mcp-remote bridge:

{
  "mcpServers": {
    "service4ai": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.service4ai.com/v1/en"]
    }
  }
}

To explore the tools without an agent, the official inspector does the job: npx @modelcontextprotocol/inspector, transport “Streamable HTTP”, then the server URL.

Exposed tools

French names (louer_humain, suivre_mission…) are accepted on both URLs.

Available tasks: bakery_line, complain, cerfa_stamp, wine_tasting, cheek_kiss, paris_parking, it_depends. Details and prices are in the catalog.

Example prompts

Once the server is connected, just talk to your agent:

• “Rent a human to stand in line at the bakery on rue des Martyrs tomorrow at 8 am.”
• “How many sighs to get a Cerfa form stamped on a Monday?”
• “Is the Friday after Ascension Day a bridge day in France?”
• “Send someone to grumble at customer service for me, intensity 4.”
• “How is the mission you started earlier going?”

The server warns the agent that humans have lunch from noon to 2 pm (Paris time) and don't work on May 1st. Well-mannered agents take it into account.

Test with curl

The server speaks JSON-RPC 2.0 over a plain POST request, with no session to open:

curl https://mcp.service4ai.com/v1/en \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"estimate_sighs","arguments":{"task":"cerfa_stamp"}}}'

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "structuredContent": {
      "task": "cerfa_stamp",
      "estimated_sighs": 4,
      "range": [3, 6],
      "confidence": "it depends"
    },
    "isError": false
  }
}

Replace tools/call with tools/list (without params) to get the list of tools and their schemas.

Troubleshooting

Something else? Write to contact@service4ai.com. A human will answer, after their coffee.