When an assistant is confidently wrong, more turns with the same model rarely help. Pulling three sources with genuinely different failure modes — another model, live web results, and human-accepted answers — is the useful shape, and the git-aware context means you do not have to paste the surrounding files yourself.
An MCP server that combines three different sources — a second model, web-grounded search, and accepted answers from real engineers — into one synthesised answer, so you can unblock a stuck conversation without switching apps.
- Multi-source synthesis: Google Gemini for an alternative model perspective, Perplexity for fresh web-grounded analysis, Stack Overflow for accepted answers
- Automatic language detection from file extensions, and code snippet extraction with clean formatting
- Git-aware context gathering that pulls in nearby files when relevant
- Markdown report generation for the solution
- One tool call, so the model gathers your code and context and returns a combined take
TypeScript and Node.js 18+. Clone, `npm install`, `npm run build`, then register `node build/index.js` in `claude_desktop_config.json`. Two keys are required: `GEMINI_API_KEY` from aistudio.google.com and `PERPLEXITY_API_KEY` from perplexity.ai. `STACK_EXCHANGE_KEY` is optional — without it the Stack Exchange calls fall back to anonymous. Restart Claude Desktop after adding the config. MIT.