What it does
Recovery Probe triggers failures such as a 503 response, a dropped connection or a broken JSON response in web and Electron applications. It observes how the UI reacts and saves HTML and JSON reports. Each faulted run is compared with a healthy run on the same screen, and each result is a pass, a failure or inconclusive.
Developers and AI coding agents can use it to find cases where a failed read turns a populated screen into an empty one, such as "No items found". Commands cover reads behind a screen, stale states, Save or Send actions, failed HTTPS connections and code that treats a failed read as no data. An MCP server lets an agent fail a read and check the screen before claiming error handling works. In Playwright tests, withFault fails matching requests, and createFaultFetch covers code that accepts a fetch function.
Running it needs Node.js 22 or newer and Chromium for web apps. Electron apps must start with a local remote debugging port, and IPC-based apps need a small development-only hook in the main process. The init and doctor commands guide setup. Reports stay on the local machine and contain no telemetry.
Limitations and requirements
Requires Node.js 22 or newer. Web apps run in Chromium. Electron apps need a local remote debugging port, and apps that load data over IPC need a development-only hook. Native HTTPS clients must use HTTPS_PROXY, and only whole connections can fail there. Android WebView support is experimental. It does not simulate a whole device going offline, WebSocket drops, token expiry, sleep or slow networks. Reports can contain endpoint paths, selectors, marker text and, with --screenshots, on-screen content. Failed recovery checks can save up to 200 characters of screen text (screenText) that is not redacted and may include personal or sensitive data, so review outputs before sharing them or passing them to an agent. Only test apps and accounts you are allowed to test, and keep debugging ports on localhost. The opt-in repair workflow calls an AI service.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 17 Nov 2026 · its page on the source
io.github.BojanKovachki/recovery-probe
Fail reads in web and Electron test apps, inspect recovery, and compare before/after evidence.
v1.1.0As shown on registry.modelcontextprotocol.io when we last read it