What problem does it solve? Writing scripts that call AiRaccoon's MCP server fails in non-obvious ways: the HTTP transport returns SSE instead of plain JSON, memory_write has no path parameter, passing context silently hides entries from search, and the ONNX embedding model is missing from build output. This Skill documents every verified pitfall and the correct client patterns. ## Core Features & Use Cases - SSE Response Parsing: Correctly parse Server-Sent Events responses and unwrap the double-nested JSON-RPC result instead of calling resp.json() directly. - Hash Pre-computation: Reproduce AiRaccoon's SHA256 path and content-hash formulas to verify written entries against search results. - Scope and Access Pitfalls: Avoid the context parameter that sets scope='custom' (invisible to memory_stats and project-scoped search) and handle the full access mode required by memory_delete_context. - Use Case: You are writing a Python ingestion script that writes documentation chunks into AiRaccoon memory and verifies them via memory_search; this Skill gives you the working client pattern and the five known failure modes with fixes. ## Quick Start Read this Skill before writing any Python, Node, or curl code that calls the AiRaccoon MCP endpoint at http://localhost:5000/mcp.
