AI and MCP
AI that reads your logs on your server, and agents that ask first.
Explain this failure names the likely cause of a failed deploy at once, without a model. With your own Anthropic key, Ask AI runs the diagnosis on the failing server, so Unrent never sees the key, the logs or the prompt. Unrent’s MCP server lets coding agents operate your services, and every change waits for a passkey.
Likely cause
Missing variable DATABASE_URL
Error: DATABASE_URL is not set
at loadConfig (/app/config.js:12)Next step
Link the database to this service, then apply the config.
Explain this failure
Every failed deploy and failed server job has an Explain button. Twelve deterministic signatures (out of memory, disk full, port in use, missing variable, image pull denied, image not found, TLS and ACME errors, DNS, a failed command, health check timeout, crash loop, early exit) name the likely cause instantly, with no key and no model.
Ask AI goes further with your own Anthropic API key. The key is sealed in your browser to each server. The agent collects the evidence, redacts it, and calls api.anthropic.com directly through an allow-listed tunnel. Unrent receives a structured result, validates it and shows it, with the cited log lines.
A suggested fix is a button (restart, roll back or start) that goes through the normal job path and its approvals.
Coming Other model providers (OpenAI-compatible, Amazon Bedrock, Google Vertex AI).
An MCP server where agents ask, and people approve.
Connect Claude Code, or any MCP client that speaks the 2026-07-28 revision, to https://mcp.unrent.dev/mcp. Sign in, pick one organisation and the scopes, and your agent gets 13 typed tools: read servers, services, deployments, logs and metrics; ask to deploy, roll back, restart, stop or start.
Every change becomes a signed job held as pending. The tool returns a link; a person reviews the job and approves it with a passkey, and only then is it delivered to the server.
claude mcp add --transport http unrent https://mcp.unrent.dev/mcp- Your agent asksdeploy · production/web
- Unrent signs the job and holds itNothing is sent to the server yet
- You approve with your passkeyYour passkey signs this exact job
- Your server checks, then actsThe agent verifies every signature
Questions
Does Unrent see my logs or my model key?
No. The key is sealed in your browser to your servers, and the diagnosis runs on the server that failed. Logs and the prompt go from your server to your Anthropic account; Unrent receives only the structured result.
Who pays for the model?
You do, through your own Anthropic account. Unrent charges nothing for AI and keeps a monthly token budget per organisation so a loop cannot run up your bill.
Can my coding agent delete things?
No tool deletes anything, and every change it asks for waits until a person approves that exact job with a passkey.
Your servers, your keys. Start with one free server.
The first server is free, then $2 per server per month, with nothing gated.