Summary
Closing the build-verify-fix loop against a real browser.
The agent that wrote the feature is a poor judge of it. Squidler exercises the app from natural-language use cases and hands back the friction it found, so the fix comes from what a user hit rather than from the implementation the agent already believes in.
What it is
A hosted MCP server for Squidler, which exercises a web app the way a person would from use cases written in natural language rather than DOM selectors. The server gives an agent the site registry and the problems those runs found, and ships built-in instructions that teach the client a test-after-feature workflow.
What you get
- Sites registered, listed, and one site's detail read back
- Every problem found, listed and summarised
- One problem opened in full
- A problem resolved or dismissed once it is dealt with
- Built-in agent guidance, so the client knows the workflow without you explaining it
Requirements
A Squidler API token from the integrations page, sent as a bearer token to the HTTP endpoint. Nothing is installed.
Setup effort
One command — npx -y mcp-remote https://mcp.squidler.io