# The MCP server

Ask Claude, ChatGPT, Cursor or VS Code about your infrastructure — "which monitors were down last night?", "where
does `mail.example.com` point?", "which domains expire this quarter?" — and let it look the answer up in InfraNest.
The assistant connects to the InfraNest **MCP server**, which reads through the same API as your scripts.

**It only reads.** Nothing an assistant does through it changes anything.

The server's address is:

```
https://api.infranest.io/mcp
```

There are two ways to connect:

- **In claude.ai and ChatGPT**, [Connect with InfraNest](ai/claude-and-chatgpt.md): add the address, sign in to
  InfraNest and approve. No token, every plan.
- **In Claude Code, Cursor, VS Code and Claude Desktop**, [use an API token](#connect-with-a-token).

The rest of this section:

- [Connect Claude and ChatGPT](ai/claude-and-chatgpt.md) — with OAuth, step by step.
- [Coding agents](ai/coding-agents.md) — an `AGENTS.md` snippet, prompts that work and safe defaults.
- [Give an assistant the docs](ai/docs-for-assistants.md) — `llms.txt` and the Markdown twins.
- [What an assistant can and cannot do](ai/safety.md) — and how to remove one.

## Connect with a token

Claude Code, Cursor, VS Code and Claude Desktop take an API token. Create one under **My account → Developer**.
Give it **read** scopes only for the areas you want the assistant to see — it cannot see more than the token can.
In your [sandbox](sandbox.md), a test token works the same way, with sample data.

The token goes in the header `Authorization: Bearer <token>`. The links below add the server to your editor and
read the token from the environment variable `INFRANEST_TOKEN`; neither link carries a token.

<McpInstall />

### Claude Code

```sh
claude mcp add --transport http infranest https://api.infranest.io/mcp \
  --header "Authorization: Bearer $INFRANEST_TOKEN"
```

### Cursor

**Add to Cursor** above, or in `~/.cursor/mcp.json` (or `.cursor/mcp.json` in a project):

```json
{
  "mcpServers": {
    "infranest": {
      "url": "https://api.infranest.io/mcp",
      "headers": { "Authorization": "Bearer ${env:INFRANEST_TOKEN}" }
    }
  }
}
```

### VS Code

**Add to VS Code** above reads `INFRANEST_TOKEN` from the environment VS Code was started from. To have VS Code ask
for the token once and keep it in its secret storage instead, put this in `.vscode/mcp.json`:

```json
{
  "inputs": [
    { "type": "promptString", "id": "infranest-token", "description": "InfraNest API token", "password": true }
  ],
  "servers": {
    "infranest": {
      "type": "http",
      "url": "https://api.infranest.io/mcp",
      "headers": { "Authorization": "Bearer ${input:infranest-token}" }
    }
  }
}
```

### Claude Desktop

The connector you add on claude.ai ([Connect Claude and ChatGPT](ai/claude-and-chatgpt.md)) is the simplest way. To
use a token instead, use the `mcp-remote` bridge (it needs Node.js) in `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "infranest": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.infranest.io/mcp", "--header", "Authorization: Bearer ${INFRANEST_TOKEN}"],
      "env": { "INFRANEST_TOKEN": "your token" }
    }
  }
}
```

## The tools

Each tool is one read in the [API reference](reference.md), and needs that read's scope. This list is generated
from the server itself.

| Tool | What it answers | What it reads | Scope |
|---|---|---|---|
| `list_organizations` | List organizations | `GET /organizations` | `members.read` |
| `status_overview` | Is everything up right now | `GET /organizations/{organization}/status` | `monitoring.read` |
| `list_servers` | List servers | `GET /organizations/{organization}/cloud-servers` | `cloud.read` |
| `get_server` | Get a server | `GET /organizations/{organization}/cloud-servers/{cloudServer}` | `cloud.read` |
| `server_attention` | What needs attention on a server | `GET /organizations/{organization}/cloud-servers/{cloudServer}/attention` | `cloud.read` |
| `server_metrics` | Servers with their latest metrics | `GET /organizations/{organization}/cloud-servers/metrics-overview` | `cloud.read` |
| `list_monitors` | List monitors | `GET /organizations/{organization}/monitoring-checks` | `monitoring.read` |
| `get_monitor` | Get a monitor | `GET /organizations/{organization}/monitoring-checks/{monitoringCheck}` | `monitoring.read` |
| `monitor_stats` | Uptime and response times of a monitor | `GET /organizations/{organization}/monitoring-checks/{monitoringCheck}/stats` | `monitoring.read` |
| `monitor_results` | Recent check results of a monitor | `GET /organizations/{organization}/monitoring-checks/{monitoringCheck}/results` | `monitoring.read` |
| `list_incidents` | List incidents | `GET /organizations/{organization}/incidents` | `monitoring.read` |
| `incident_timeline` | Timeline of an incident | `GET /organizations/{organization}/incidents/{incident}/updates` | `monitoring.read` |
| `list_dns_zones` | List DNS zones | `GET /organizations/{organization}/dns-zones` | `dns.read` |
| `list_dns_records` | List the records in a DNS zone | `GET /organizations/{organization}/dns-zones/{dnsZone}/dns-records` | `dns.read` |
| `search_dns` | Search DNS records across all zones | `GET /organizations/{organization}/dns/search` | `dns.read` |
| `list_domains` | List domains | `GET /organizations/{organization}/domains` | `domains.read` |
| `list_certificates` | List certificates | `GET /organizations/{organization}/certificates` | `monitoring.read` |
| `list_maintenance_windows` | List maintenance windows | `GET /organizations/{organization}/maintenance-windows` | `monitoring.read` |
| `search_audit_log` | Search the audit log | `GET /organizations/{organization}/audit-logs` | `audit-log.read` |
| `uptime_report` | Monthly uptime report | `GET /organizations/{organization}/uptime-report` | `monitoring.read` |
| `read_api` | Read any public API endpoint | any read in the [API reference](/reference) | the scope of what it reads |

When a token reaches more than one organization, the assistant asks `list_organizations` first and names the one
it means. Connected with InfraNest, it reaches the one organization you picked. A long answer is cut off at
80,000 characters, with a note telling the assistant how to narrow it.

## Things to ask

- "Is anything down right now?"
- "Which monitors were down last night, and for how long?"
- "Where does `mail.example.com` point, and when was that record last changed?"
- "Which domains and certificates expire in the next 60 days?"
- "Which servers are running out of disk space?"
- "Who changed the DNS for `shop.example.com` this week?"
- "What was our uptime last month, per monitor?"

## What it costs

Nothing extra, on every plan — Free included. Each tool call is an API read, and counts against the same
[rate limit](limits.md#rate-limits) as any other request with that token.

**Connect with InfraNest** works on every plan. **Connecting with a token** needs a plan that includes API tokens
(Pro or higher), or a test token in your [sandbox](sandbox.md). See [Limits](limits.md#what-each-plan-includes).

When something does not work, see [Troubleshooting](troubleshooting.md#assistant-and-mcp-connection-problems).