To run localy:
1. Download ollama and two recommended models: gpt-oss:20b-cloud and nomic-embed-text

2. Specify env variables
Defaults for ollama run:
    USER_TOKEN=<allure-api-key>
    EMBEDDINGS_PROVIDER=ollama
    LLM_PROVIDER=ollama
    OLLAMA_BASE_URL=http://localhost:11434
    OLLAMA_MODEL=gpt-oss:20b-cloud
    BASE_URL=<allure-url>

3. run python main.py
5. run streamlit run client.py or use Chrome pluggin to interact with the service

Allure MCP server integration (separate from legacy API logic):
1. Install and build https://github.com/armanayvazyan/allure-testops-mcp
2. Set env vars for MCP server:
    ALLURE_MCP_COMMAND=node
    ALLURE_MCP_ARGS="/path/to/allure-testops-mcp/dist/index.js"
    ALLURE_MCP_CWD=/path/to/allure-testops-mcp
    ALLURE_TESTOPS_URL=<allure-url>
    ALLURE_TOKEN=<allure-api-key>
    ALLURE_PROJECT_ID=<id>
    ALLURE_TREE_ID=<id>
    ALLURE_PROJECT_NAME=<project-name>
3. Use MCP endpoints/tools:
    /get_test_case_mcp
    /load_test_cases_from_allure_mcp
    /create_test_case_mcp
    /rebuild_fts_index_mcp

FTS index (lightweight search):
- Build index via /rebuild_fts_index_mcp (or MCP tool rebuild_fts_index_mcp)
- Generation uses FTS ids first, then fetches full details via MCP