Delegate RAG and query enrichment to vtk-mcp

[vicentebolea]

Summary

• Removes all local RAG infrastructure (ChromaDB, llama_index, sentence_transformers, tree_sitter, rag-components submodule, data/examples, db/)
• Delegates vector search and query enrichment to vtk-mcp via a new vtk_mcp_client.py HTTP client
• --rag now requires --mcp-url pointing to a running vtk-mcp server
• Import validation and auto-fixing of generated code via vtk-mcp (always active when --mcp-url is set)

How it works

When --mcp-url is set, a single VTKMCPClient instance is created per query and used for:

1. Context enrichment (when --rag): hybrid vector search over VTK examples + Python API docs for relevant VTK classes
2. Import fixing (always): validates and auto-corrects VTK import statements in the generated code

Changes

• vtk_mcp_client.py: new HTTP JSON-RPC client wrapping vtk-mcp tools
• client.py: removed ChromaDB code paths; added mcp_url field
• cli.py: replaced --collection/--database with --mcp-url
• pyproject.toml: removed all RAG deps; added requests; removed rag-chat, vtk-build-rag, vtk-test-rag entry points
• Deleted: rag_chat_wrapper.py, build_rag_db.py, test_rag.py, rag-components/ submodule, data/, db/

Usage

# RAG-enhanced generation
vtk-prompt "create a red sphere" --rag --mcp-url http://localhost:8000 -t $API_KEY

# Import fixing only (no RAG context)
vtk-prompt "create a cube" --mcp-url http://localhost:8000 -t $API_KEY

# No MCP (baseline)
vtk-prompt "create a cone" -t $API_KEY

Test plan

• All existing tests pass (45 passed, 14 skipped)
• RAG-enhanced generation works with a running vtk-mcp server
• Import statements in generated code are validated and fixed
• Baseline generation without --mcp-url still works