Port docs-mcp-template upgrades (eval, citations, chunking, prefixes) (#13)

Co-authored-by: claude <[email protected]>

rag/embeddings.py

@@ -22,8 +22,14 @@ import logging
 import time
 from typing import Any
 
-import httpx
-from chromadb import EmbeddingFunction, Documents, Embeddings
+try:
+    import httpx
+    from chromadb import Documents, EmbeddingFunction, Embeddings
+except ImportError:
+    httpx = None  # type: ignore[assignment]
+    EmbeddingFunction = object  # type: ignore[misc,assignment]
+    Documents = list  # type: ignore[misc,assignment]
+    Embeddings = list  # type: ignore[misc,assignment]
 
 log = logging.getLogger(__name__)
 
@@ -41,6 +47,21 @@ def _resolve_urls() -> list[str]:
 OLLAMA_URLS = _resolve_urls()
 EMBED_MODEL = os.environ.get("EMBED_MODEL", "nomic-embed-text")
 EMBED_DIM = int(os.environ.get("EMBED_DIM", "768"))
+# nomic asymmetric prefixes — embed time only. Empty string disables.
+EMBED_QUERY_PREFIX = os.environ.get("EMBED_QUERY_PREFIX", "search_query: ")
+EMBED_DOC_PREFIX = os.environ.get("EMBED_DOC_PREFIX", "search_document: ")
+
+
+def apply_prefix(text: str, prefix: str) -> str:
+    if not prefix:
+        return text
+    return prefix + text
+
+
+def embed_texts(texts: list[str], *, prefix: str, ef: Any | None = None) -> list[list[float]]:
+    """Embed texts after applying prefix. Callers store the original strings."""
+    fn = ef if ef is not None else embedding_function()
+    return list(fn([apply_prefix(t, prefix) for t in texts]))
 
 
 class OllamaEmbeddings(EmbeddingFunction):