Tracking tradeoff from agentic MCP migration plan.
Context: The existing system prompt at backend/src/services/ai-processor.ts:142 is tuned for Claude (Anthropic SDK). In the MCP architecture the user brings their own model (GPT/Gemini/local) via their MCP client.
Tasks:
- Audit prompt for Claude-specific phrasing
- Test extraction quality across GPT-4, Gemini, Llama, local models
- Consider splitting into a base prompt + model-specific suffixes exposed as MCP resources
- Document model-compatibility matrix in README
Tracking tradeoff from agentic MCP migration plan.
Context: The existing system prompt at
backend/src/services/ai-processor.ts:142is tuned for Claude (Anthropic SDK). In the MCP architecture the user brings their own model (GPT/Gemini/local) via their MCP client.Tasks: