# NotWorking [Health: Active]

**Category:** 📊 Monitoring  
**Repository:** https://github.com/RyanNSJ/notworking  
**GitHub Stars:** 2  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/notworking

## Description
Downdetector for AI agents: is a site, MCP server or skill failing for everyone? Check and report.

## Tools
Capabilities this server exposes over MCP:

- **check_status** — Is it down or blocking agents for everyone, or just you? Use when a site, URL, MCP server or skill isn't working: down, unreachable or an outage, a timeout, blocked (access denied, 403, 429 rate limited, a Cloudflare 'Just a moment' challenge, a CAPTCHA), a login loop, an MCP server that won't connect or a tool that errors, or a skill that fails. Look up a service by the URL you used, an access-path id (clawhub:<owner>/<slug>, skills.sh:<owner>/<repo>/<skill>, an MCP Registry name like io.github.<owner>/<server>) or a service name. Returns a one-sentence summary, the service, every listed access path with a description and failure reports from the last hour (and the last 24 hours as context), the what_failed values and how to report. Unlisted targets return their own counts and status, with listed: false.
- **report** — Report that an access path failed for you. Use the access-path id from check_status, or the URL you used. Only report failures that came from the service, not ones caused by your own tools, permissions or settings (like a fetch tool that refuses a domain). Anonymous. Never include personal data, booking details, credentials or full URLs with query strings. target is which access path failed; what_failed is how, from this list (the same for every service): site_unreachable: The site or server could not be reached at all, for example a DNS error, a timeout or a 5xx server error.; bot_block: The service refused the agent: an access-denied page, a 403 or 429 error, a bot wall or a challenge page.; captcha: A CAPTCHA or 'are you human?' check stopped the agent.; login_auth: Logging in or authenticating failed, for example a login loop, a new login step or rejected credentials.; api_error: An API the agent used returned an error or an unexpected response.; mcp_error: An MCP server failed to connect, initialise, list its tools or run a tool.; skill_failed: A skill could not be installed, or it ran but errored or produced a wrong result.; payment: A payment or checkout step failed or was blocked.; specific_page: One particular page or step failed while the rest of the service seemed to work.; other: Something else went wrong.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What NotWorking MCP server does

NotWorking MCP server provides outage and access-failure checks for websites, MCP servers, and agent skills. It is intended for cases where a service may still work for people but fails when accessed by an agent. The service collects reports from agents and a daily canary, then exposes recent status information through MCP.

The catalogue can identify services by the URL used, a service name, or an access-path identifier. Supported identifier formats include `clawhub:<owner>/<slug>`, `skills.sh:<owner>/<repo>/<skill>`, and MCP Registry names such as `io.github.<owner>/<server>`.

This server belongs in the Monitoring category when an agent needs evidence about whether a failure is local or shared. The catalogue is described as small and reports are still accumulating, so an absent or sparse result is not proof that a service is healthy.

## How it works

The `check_status` tool returns a short status summary, the matched service, its listed access paths, descriptions, and failure reports. Results cover reports from the previous hour, with the previous 24 hours included for context. The response also includes the recorded `what_failed` values and instructions for submitting a report. If the target is not in the catalogue, the response returns its own counts and status with `listed: false`.

The `report` tool records an anonymous failure against an access-path identifier or the URL used. Reports should describe a failure caused by the service itself, rather than a restriction from the agent’s own permissions, settings, or fetch tools. Failure categories include unreachable services, bot blocks, CAPTCHAs, authentication failures, API errors, MCP errors, skill failures, payment problems, page-specific failures, and other issues.

## Setup and configuration

NotWorking MCP server is available at `https://notworking.io/mcp` as a streamable HTTP MCP endpoint and does not require sign-in. Any MCP client that can connect to that endpoint can use it. The README also documents installation through Claude Code, Codex, Gemini CLI, pi, and the skills CLI. The repository includes a local development path using `uv sync` and `uv run agentdown serve`, which serves on `127.0.0.1:8000` after migrating the database.

## Tools and capabilities

- `check_status`: inspect a service, URL, access path, or service name and compare recent reports.
- `report`: submit an anonymous report for a failed access path.
- Classify failures using the shared `what_failed` values across services.
- Review alternate listed access paths and their descriptions.

## Limitations and notes

NotWorking MCP server should not receive personal data, booking details, credentials, or full URLs containing query strings. The `report` tool is for service-caused failures, not problems created by local tools, permissions, or settings. The catalogue and report history are early-stage, and unlisted targets are explicitly marked rather than treated as catalogue-confirmed services.

The code is licensed under MIT. The data and catalogue use CC BY 4.0 and require credit to “NotWorking (notworking.io)” when applicable.

_Full upstream README: https://allmcps.com/mcp/notworking/readme_