Update hf_api_tool.py with token passthrough support

hf_api_tool.py

@@ -13,10 +13,24 @@ DEFAULT_TIMEOUT_SEC = 30
 
 
 def _load_token() -> str | None:
+    # Check for request-scoped token first (when running as MCP server)
+    # This allows clients to pass their own HF token via Authorization header
+    try:
+        from fast_agent.mcp.auth.context import request_bearer_token
+
+        ctx_token = request_bearer_token.get()
+        if ctx_token:
+            return ctx_token
+    except ImportError:
+        # fast_agent.mcp.auth.context not available
+        pass
+
+    # Fall back to HF_TOKEN environment variable
     token = os.getenv("HF_TOKEN")
     if token:
         return token
 
+    # Fall back to cached huggingface token file
     token_path = Path.home() / ".cache" / "huggingface" / "token"
     if token_path.exists():
         token_value = token_path.read_text(encoding="utf-8").strip()
@@ -76,6 +90,7 @@ def hf_api_request(
     params: dict[str, Any] | None = None,
     json_body: dict[str, Any] | None = None,
     max_results: int | None = None,
+    offset: int | None = None,
 ) -> dict[str, Any]:
     """
     Call the Hugging Face Hub API (GET/POST only).
@@ -86,6 +101,7 @@ def hf_api_request(
         params: Optional query parameters.
         json_body: Optional JSON payload for POST requests.
         max_results: Max results when response is a list (defaults to HF_MAX_RESULTS).
+        offset: Client-side offset when response is a list (defaults to 0).
 
     Returns:
         A dict with the response data and request metadata.
@@ -119,9 +135,7 @@ def hf_api_request(
             status_code = response.status
     except HTTPError as exc:
         error_body = exc.read().decode("utf-8", errors="replace")
-        raise RuntimeError(
-            f"HF API error {exc.code} for {url}: {error_body}"
-        ) from exc
+        raise RuntimeError(f"HF API error {exc.code} for {url}: {error_body}") from exc
     except URLError as exc:
         raise RuntimeError(f"HF API request failed for {url}: {exc}") from exc
 
@@ -132,7 +146,9 @@ def hf_api_request(
 
     if isinstance(payload, list):
         limit = max_results if max_results is not None else _max_results_from_env()
-        payload = payload[: max(limit, 0)]
+        start = max(offset or 0, 0)
+        end = start + max(limit, 0)
+        payload = payload[start:end]
 
     return {
         "url": url,