> For the complete documentation index, see [llms.txt](https://docs.monogoto.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.monogoto.io/developer/mcp-connectivity-for-ai-agents-new.md).
# MCP - Connectivity for AI agents (New!)
The Monogoto MCP server exposes the platform to AI agents through the Model Context Protocol. Instead of writing REST calls, you point an agent at your account and ask in plain language - "which Things in the Berlin group went quiet this week", "show me the top data consumers on Vodafone", "suspend ICCID 8944...". The agent picks the tools and makes the calls.
Same platform as the REST API, same permissions, same audit trail. Over 120 tools cover Things, reports, NetFlow analysis, orders, campaigns, packet capture, users, tags, integrations, and payments.
Looking for the REST API instead? See API or go straight to the [interactive API library](https://developer.monogoto.io/overview).
#### Endpoint
```
https://mcp.monogoto.io/v1/mcp
```
Streamable HTTP, MCP protocol `2025-06-18`. There is no separate SSE endpoint - streaming is negotiated on the same URL with `Accept: application/json, text/event-stream`.
#### Authentication
The server is an OAuth 2.0 protected resource with dynamic client registration, so most clients need nothing beyond the URL. On first connection your client registers itself, opens a browser for you to sign in to Monogoto, and receives its own credentials. Access tokens are short-lived and your client refreshes them in the background.
You never put a password or a token in a config file.
#### Check your client first
Your AI client must support **remote MCP servers** reached over HTTPS. Clients that only support locally-run servers cannot connect to this endpoint. Check your client's own documentation before you start, as support and setup differ between products and change often.
On Claude specifically: custom connectors are available on Free, Pro, Max, Team, and Enterprise plans, and free accounts are limited to one custom connector. **If your organisation is on a Team or Enterprise plan, an Owner must add the connector at the organisation level before you can use it.** Individual members cannot add it themselves. Plan your evaluation accordingly, or ask your Claude workspace administrator up front.
#### Connect your client
**Claude Desktop and claude.ai** 
Settings, then Connectors, then Add custom connector. Paste `https://mcp.monogoto.io/v1/mcp` and follow the sign-in prompt in your browser.
On Team or Enterprise plans, an Owner first adds it under Organisation settings, then each member connects their own Monogoto account.
**Claude Code**
bash
```bash
claude mcp add --transport http monogoto https://mcp.monogoto.io/v1/mcp
```
Run `/mcp` in Claude Code to complete authentication.
**Cursor** 
`.cursor/mcp.json`:
json
```json
{
"mcpServers": {
"monogoto": {
"url": "https://mcp.monogoto.io/v1/mcp"
}
}
}
```
Cursor shows a "Needs login" prompt on first use. Click it to authorize.
**VS Code** 
**/ Copilot** 
`.vscode/mcp.json`, same shape as Cursor.
**Any other MCP client**
Windsurf, Cline, Zed, Goose, and n8n all accept a remote server URL and handle the OAuth handshake. Use the Cursor block as the template.
**Your own application**
Call the endpoint directly with JSON-RPC 2.0 over POST, sending `Accept: application/json, text/event-stream`.
An unauthorized request returns `401` with a `WWW-Authenticate` challenge that advertises the authorization server, so you can discover the OAuth endpoints rather than hard-coding them. From there, register a client at `https://mcp.monogoto.io/register`, run an authorization-code flow with PKCE at `/authorize`, and exchange the code at `/token`.
Discover tools with `tools/list` after connecting, and again after any reconnect - tool names and schemas can change independently of the REST API specification. For `tools/call`, `params` must carry both `name` and an `arguments` object; pass `"arguments": {}` for tools with no required fields.
Alternatively, use the Claude Messages API MCP connector and pass the server URL.
#### What the tools cover
| Domain | What agents can do |
| --------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Things | Status, location, real-time connectivity, signalling events, identity and PLMN, lifetime usage, QoS tests, SMS and bulk SMS, tags, activate and suspend |
| Reports | Usage, insights, top consumers, groups, networks and identities, billing reports |
| NetFlow | Destinations, services, protocols, TCP flags, enriched records, network diagrams |
| Orders | Product plans, order creation, shipment tracking |
| Packet Capture | Create, monitor, download, and delete captures |
| SIM Profiles | SoftSIM via Nordic, eSIM profiles |
| Campaigns | Progress, membership, exports |
| Users, Tags, Groups | Full lifecycle and organisation |
| Integrations | Email, webhook, and Zapier |
| Payments, Inbox, Audit Logs, Rate Plans | Invoices, notifications, audit trail, plan listing |
The tools you see depend on your permissions - administrators get brand management tools that other roles don't. Call `tools/list` on a connected session for the authoritative list for your account.
#### Before you connect an agent
There is no sandbox. These tools run against your live account, and several are destructive or spend money:
* `change-thing-status` suspends or activates real SIMs
* `send-bulk-sms` messages up to 100 devices in one call
* `create-order` places a real hardware order
* `delete-user`, `delete-tag`, `delete-integration`, `delete-capture`, `cancel-esim` are irreversible
The server honours the permissions of the account you authorize with. Sign in with a role scoped to the minimum the task needs rather than an admin account, and point early experiments at a dedicated test group. Every action is queryable through `list-audit-logs`.