AI Agents (MCP)
Dashboard.PT has an MCP (Model Context Protocol) server, the standard that lets AI agents query other services. Once connected, you can ask the agent, for example, which cronjobs are failing or when a site's SSL certificate expires, and it fetches the answer from your account.
Overview
What to know
- PLUS and GOLD plans. The connection is only available on these plans. If the account moves to another plan, the agent loses access until it returns to one of these.
- Read-only. The agent can look up data, but it cannot create, change, run or delete anything.
- Your account only. The agent only sees the data of the account you authorized.
- Server address:
https://dashboard.pt/api/mcp.php
⚠ Privacy: whatever the agent looks up is sent to the AI provider you chose, which may process it outside Europe under its own terms. This includes cronjob and monitor names, URLs, graph values and your reminder notes. The connection is optional: it only happens if you authorize it.
Connect an agent
Menu names may change, as they depend on each application.
Claude (web and Desktop)
- In Claude's settings, open connectors and add a custom connector.
- Enter the address
https://dashboard.pt/api/mcp.php. - Log in to Dashboard.PT if asked.
- Check the return address claude.ai on the authorization page and click Authorize.
Claude Code
Add the server, then authorize it with the /mcp command inside Claude Code:
claude mcp add --transport http dashboard-pt https://dashboard.pt/api/mcp.php
ChatGPT and other clients
The server follows the MCP standard, with Streamable HTTP transport and OAuth 2.1, so it may work with other compatible clients. ChatGPT only lets you add your own servers in developer mode, on some plans. We have not yet confirmed the connection with ChatGPT.
What the agent can look up
| Tool | What it returns |
|---|---|
list_cronjobs | Cronjobs, those in error first. Can filter by status and search by name |
get_cronjob_executions | Latest executions of a cronjob: HTTP status, duration and the start of the response text |
list_monitors | Monitors: status, uptime, and SSL and domain expiry |
list_dashboards | Dashboards and the number of graphs in each |
list_graphs | Graphs, from all dashboards or from one |
get_graph_data | Data points of a graph, with an optional date range |
list_alerts | Alert rules and when they last fired |
list_reminders | Reminders, pending only by default |
Security
- Authorization is always requested. The page shows the return address, because the application supplies its own name. If the address is not what you expect, decline.
- Hidden credentials. In URLs, the values of parameters such as
token,keyorsecretappear as***. Cronjob request headers and bodies are never sent, only the header names. - A limit of this protection. A secret in the URL path, or written in a cronjob name or reminder title, is not detected.
- Third-party text. Cronjob responses come from other sites. The server marks them as untrusted, but malicious text can try to influence the agent, so review what it suggests.
Disconnect an agent
On the API Keys page, the Connected applications section shows each connection, with its date and last use. Click Disconnect to revoke access immediately. To use it again, you have to authorize it again.
Technical summary
| Item | Details |
|---|---|
| Plans | PLUS and GOLD. On the others, the server replies 403 |
| Transport | Streamable HTTP (POST only, no session) |
| Protocol versions | 2025-03-26, 2025-06-18 and 2025-11-25 |
| Authorization | OAuth 2.1 with PKCE S256 and dynamic client registration (RFC 7591) |
| Token lifetime | Access: 1 hour. Refresh: 30 days, replaced on each use |
| Hourly limit | Tool calls, added to API requests: 1000 on the PLUS and GOLD plans |
| OAuth alternative | API key in the Authorization: Bearer header, for clients that let you set headers |
Want to try it? The free plan has no time limit.
Create a free account