MCP server

Snapback

Diagnose why an AI agent failed and get the verified fix instantly. Free, no token.

Endpoint: Answering
Works in: 20 of 21 clients
Tools: 23
Last checked: Sep 25, 2026

Does Snapback work in your MCP client?

Registry name io.github.ra1labsworkx-wq/snapback, endpoint https://api.snapback.sh/mcp.

Claude Codeworks
Claude Desktopworks
Claude.aiworks
Claude API MCP connectorworks
ChatGPTworks
OpenAI Responses APIworks
Codexworks
Gemini CLIworks
Cursorworks
VS Code (GitHub Copilot)works
Windsurf (Devin Desktop)works
Zedworks
JetBrains AI Assistantworks
Junieunknownno source confirms Junie supports Streamable HTTP
Clineworks
Continueworks
Gooseworks
Lovableworks
Replit Agentworks
Bolt.newworks
v0works

How clients sign in

No credentials: it answered lurq's probe without any.

Tools

diagnose_tracediagnose_batchget_verdictpreflightsubmit_feedbackget_request_statusmy_usagemy_impactwhat_others_didreport_outcomerecommend_failovercascade_rootsuggest_budget_recoveryrequest_patterndetect_loopsession_startsession_stepsession_endagent_memorybudget_guardconvert_tracesearch_docsdiagnose_infra_error

Get the setup for your client from your agent

This page is a daily snapshot. lurq's connect_check answers for the client you are wiring this server into, with the exact steps and the config to paste in that client's own format, and lurq mcp-pin tells you when the server changes. One command connects Claude Code, Cursor, VS Code and other agents:

npx lurqrun

Free to start. How connect_check works

From lurq's credential-free probe of this server's endpoint on Sep 25, 2026, and each client's own documentation. This page refreshes every 168 hours. How the probe works, and how maintainers opt out. Source repository.