Skip to content
Docs
AI agents (MCP)

Connecting AI agents

MCP in Multogram: how to connect Claude Desktop, Claude Code, Cursor, VS Code, Codex and other AI clients to the application. Where to get the mcp.exe MCP server, downloading the archive, setting up the client and checking the connection.

MCP (Model Context Protocol) is an open standard for connecting AI agents to programs. Through the Multogram MCP server your AI client — Claude Desktop, Claude Code, Cursor, VS Code, Codex or any other with MCP support — works with the running application: it answers from the documentation, shows tasks, accounts and proxies, goes through logs and helps set up modes.

MCP is for those who already work in their own AI client. The chat inside the Multogram window is the separate AI assistant; nothing needs to be connected for it.

What you need

  • Multogram is running under the same Windows user account as the AI client.
  • You are signed in to your Multogram account. Access is confirmed by the Multogram server, so you need an internet connection.
  • An AI client with MCP support (connecting over stdio — this is how all popular clients work).

MCP is available on all plans, including Free; the plan determines what exactly the agent can do — see What the agent can do. There is nothing to turn on in the application: the tools appear for the agent by themselves while Multogram is running and you are signed in, and disappear when you sign out or close the application.

Where to get the MCP server

The MCP server is the mcp.exe program. The AI client starts it itself, and it passes requests to the open Multogram. The server needs no addresses, ports or keys: it finds the running application by itself.

  • In the Multogram folder. The server is installed together with the application: mcp\mcp.exe in the folder where Multogram is installed.
  • As a separate archive. If there is no mcp folder or you need the server somewhere else, download the archive and unpack it into any folder. The path for the client is <folder>\mcp\mcp.exe.

Multogram MCP server

ZIP archive · Windows 10/11 · 64-bit

Download

The server and the application come from one release

Multogram accepts connections only from the MCP server of its own release and earlier ones. If the agent reports that the versions don't match or that Multogram rejected the connection, update Multogram and take mcp.exe from its folder.

Connecting

Find the path to mcp.exe — for example, C:\Program Files\Multogram\mcp\mcp.exe.

Add the server to the client's settings — examples for popular clients are below. In JSON and TOML the path is easier to write with forward slashes: C:/Program Files/Multogram/mcp/mcp.exe.

Restart the client (or reconnect the server) and ask the agent to call get_access_status. It answers whether Multogram is running, whether you are signed in, what the plan unlocks and what to do if there is no access.

Settings → Developer → Edit Config — the claude_desktop_config.json file opens:

claude_desktop_config.json
{
  "mcpServers": {
    "multogram": { "command": "C:/Program Files/Multogram/mcp/mcp.exe" }
  }
}

With one command:

claude mcp add multogram -- "C:\Program Files\Multogram\mcp\mcp.exe"

Or for a single project — a .mcp.json file in its folder, in the same format as for Claude Desktop.

The .cursor/mcp.json file in the project, or ~/.cursor/mcp.json for all projects:

mcp.json
{
  "mcpServers": {
    "multogram": { "command": "C:/Program Files/Multogram/mcp/mcp.exe" }
  }
}

The .vscode/mcp.json file in the project folder:

.vscode/mcp.json
{
  "servers": {
    "multogram": { "type": "stdio", "command": "C:/Program Files/Multogram/mcp/mcp.exe" }
  }
}

The ~/.codex/config.toml file:

config.toml
[mcp_servers.multogram]
command = "C:/Program Files/Multogram/mcp/mcp.exe"

Other clients connect the same way: the command is the path to mcp.exe, with no arguments.

What to ask the agent

  • "How do I set up proxies for accounts in Multogram?"
  • "Show the tasks that are running now and the latest errors in the log"
  • "Why did the mailer task stop? Go through the log and find the cause in the documentation"
  • "How many accounts have an invalid session?"
  • "Open the Proxies tab"
  • "Add the accounts from the folder D:\Sessions, check them and give each its own working proxy" — on any plan
  • "Add these proxies and check which of them work" — on any plan
  • "Fill in the Mailer settings: delay between messages 30–60 seconds" — on the Pro and Max plans
  • "Start a mailing from the active accounts, show me what will run first" — on the Pro and Max plans

You don't need to explain how Multogram works to the agent: together with the tools it gets a guide to the application and answers from this documentation, not from memory.

Security

  • The agent works with your Multogram live. Connect only clients and agents you trust.
  • You confirm the changes. Before changes, task launches and removals Multogram shows a confirmation window — see Confirmations.
  • Secrets are out of the agent's reach. It doesn't read or fill in API keys and passwords — you enter them.
  • The server checks access. Access lasts while you are signed in on this computer and closes when you sign out.
  • mcp.exe stores nothing. It holds no keys or passwords: it only passes requests to the application running on the same computer.

Common questions

Do I need a paid plan for MCP?

No. On Free the agent answers from the documentation, shows tasks, accounts, proxies and logs, helps with navigation, and adds and checks accounts and proxies, assigns proxies and creates templates — what you do by hand. It can set up modes and start tasks on Pro and Max — see What the agent can do.

Does MCP spend AI credits?

No. The model runs in your AI client and is paid for there. Multogram's AI credits are spent only by the built-in AI assistant.

Can I connect several agents?

Yes, within your plan's limit of simultaneously running sessions — see Limits.

The agent doesn't see Multogram's tools — what should I do?

Make sure Multogram is running and you are signed in, then ask the agent to call get_access_status. Other cases are in If the agent won't connect.

Updated

CookiesPrivacy Policy

We use essential cookies to run this site. With your consent, we will also use analytics cookies to improve it and marketing cookies to credit whoever referred you. You can change your choice at any time.