mcp: support http(s) URLs for spec_path in OpenAPI MCP loader

litellm-proxy-extras/litellm_proxy_extras/migrations/20260209085821_add_verificationtoken_indexes/migration.sql

@@ -0,0 +1,8 @@
+-- CreateIndex
+CREATE INDEX "LiteLLM_VerificationToken_user_id_team_id_idx" ON "LiteLLM_VerificationToken"("user_id", "team_id");
+
+-- CreateIndex
+CREATE INDEX "LiteLLM_VerificationToken_team_id_idx" ON "LiteLLM_VerificationToken"("team_id");
+
+-- CreateIndex
+CREATE INDEX "LiteLLM_VerificationToken_budget_reset_at_expires_idx" ON "LiteLLM_VerificationToken"("budget_reset_at", "expires");

litellm-proxy-extras/litellm_proxy_extras/schema.prisma

@@ -310,6 +310,16 @@ model LiteLLM_VerificationToken {
     litellm_budget_table LiteLLM_BudgetTable?   @relation(fields: [budget_id], references: [budget_id])
     litellm_organization_table LiteLLM_OrganizationTable?   @relation(fields: [organization_id], references: [organization_id])
     object_permission LiteLLM_ObjectPermissionTable?   @relation(fields: [object_permission_id], references: [object_permission_id])
+
+    // SELECT COUNT(*) FROM (SELECT "public"."LiteLLM_VerificationToken"."token" FROM "public"."LiteLLM_VerificationToken" WHERE ("public"."LiteLLM_VerificationToken"."user_id" = $1 AND ("public"."LiteLLM_VerificationToken"."team_id" IS NULL OR "public"."LiteLLM_VerificationToken"."team_id" <> $2)) OFFSET $3 ) AS "sub"
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE "public"."LiteLLM_VerificationToken"."user_id" = $1 OFFSET $2
+    @@index([user_id, team_id])
+
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE "public"."LiteLLM_VerificationToken"."team_id" = $1 OFFSET $2
+    @@index([team_id])
+
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE (("public"."LiteLLM_VerificationToken"."expires" IS NULL OR "public"."LiteLLM_VerificationToken"."expires" > $1) AND "public"."LiteLLM_VerificationToken"."budget_reset_at" < $2) OFFSET $3
+    @@index([budget_reset_at, expires])
 }
 
 // Audit table for deleted keys - preserves spend and key information for historical tracking

litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py

@@ -3,6 +3,8 @@
 """
 
 import json
+import asyncio
+import os
 from pathlib import PurePosixPath
 from typing import Any, Dict, Optional
 from urllib.parse import quote
@@ -45,8 +47,36 @@ def _sanitize_path_parameter_value(param_value: Any, param_name: str) -> str:
 
 
 def load_openapi_spec(filepath: str) -> Dict[str, Any]:
-    """Load OpenAPI specification from JSON file."""
-    with open(filepath, "r") as f:
+    """
+    Sync wrapper. For URL specs, use the shared/custom MCP httpx client.
+    """
+    try:
+        # If we're already inside an event loop, prefer the async function.
+        asyncio.get_running_loop()
+        raise RuntimeError(
+            "load_openapi_spec() was called from within a running event loop. "
+            "Use 'await load_openapi_spec_async(...)' instead."
+        )
+    except RuntimeError as e:
+        # "no running event loop" is fine; other RuntimeErrors we re-raise
+        if "no running event loop" not in str(e).lower():
+            raise
+    return asyncio.run(load_openapi_spec_async(filepath))
+
+async def load_openapi_spec_async(filepath: str) -> Dict[str, Any]:
+    if filepath.startswith("http://") or filepath.startswith("https://"):
+        client = get_async_httpx_client(llm_provider=httpxSpecialProvider.MCP)
+        # NOTE: do not close shared client if get_async_httpx_client returns a shared singleton.
+        # If it returns a new client each time, consider wrapping it in an async context manager.
+        r = await client.get(filepath)
+        r.raise_for_status()
+        return r.json()
+
+    # fallback: local file
+    # Local filesystem path
+    if not os.path.exists(filepath):
+        raise FileNotFoundError(f"OpenAPI spec not found at {filepath}")
+    with open(filepath, "r", encoding="utf-8") as f:
         return json.load(f)
 
 

litellm/proxy/schema.prisma

@@ -308,6 +308,16 @@ model LiteLLM_VerificationToken {
     litellm_budget_table LiteLLM_BudgetTable?   @relation(fields: [budget_id], references: [budget_id])
     litellm_organization_table LiteLLM_OrganizationTable?   @relation(fields: [organization_id], references: [organization_id])
     object_permission LiteLLM_ObjectPermissionTable?   @relation(fields: [object_permission_id], references: [object_permission_id])
+
+    // SELECT COUNT(*) FROM (SELECT "public"."LiteLLM_VerificationToken"."token" FROM "public"."LiteLLM_VerificationToken" WHERE ("public"."LiteLLM_VerificationToken"."user_id" = $1 AND ("public"."LiteLLM_VerificationToken"."team_id" IS NULL OR "public"."LiteLLM_VerificationToken"."team_id" <> $2)) OFFSET $3 ) AS "sub"
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE "public"."LiteLLM_VerificationToken"."user_id" = $1 OFFSET $2
+    @@index([user_id, team_id])
+
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE "public"."LiteLLM_VerificationToken"."team_id" = $1 OFFSET $2
+    @@index([team_id])
+
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE (("public"."LiteLLM_VerificationToken"."expires" IS NULL OR "public"."LiteLLM_VerificationToken"."expires" > $1) AND "public"."LiteLLM_VerificationToken"."budget_reset_at" < $2) OFFSET $3
+    @@index([budget_reset_at, expires])
 }
 
 // Audit table for deleted keys - preserves spend and key information for historical tracking

litellm/responses/litellm_completion_transformation/streaming_iterator.py

@@ -88,6 +88,8 @@ def __init__(
         self._pending_tool_events: List[BaseLiteLLMOpenAIResponseObject] = []
         self._tool_output_index_by_call_id: dict[str, int] = {}
         self._tool_args_by_call_id: dict[str, str] = {}
+        self._tool_call_id_by_index: dict[int, str] = {}
+        self._ambiguous_tool_call_indexes: set[int] = set()
         self._next_tool_output_index: int = 1  # output_index=0 reserved for the message item
         self._final_tool_events_queued: bool = False
         self._sequence_number: int = 0  
@@ -111,6 +113,19 @@ def _get_or_assign_tool_output_index(self, call_id: str) -> int:
         self._tool_output_index_by_call_id[call_id] = idx
         return idx
 
+    def _normalize_tool_call_index(self, tool_call: object) -> Optional[int]:
+        idx_raw = (
+            tool_call.get("index")
+            if isinstance(tool_call, dict)
+            else getattr(tool_call, "index", None)
+        )
+        if idx_raw is None:
+            return None
+        try:
+            return int(idx_raw)
+        except (TypeError, ValueError):
+            return None
+
 
     def _is_reasoning_end(self, chunk):
         delta = chunk.choices[0].delta
@@ -143,10 +158,28 @@ def _queue_tool_call_delta_events(self, tool_calls: object) -> None:
             return
 
         for tc in tool_calls:
+            tc_index = self._normalize_tool_call_index(tc)
             call_id_raw = tc.get("id") if isinstance(tc, dict) else getattr(tc, "id", None)
-            if not call_id_raw:
+            call_id = ""
+
+            if call_id_raw:
+                call_id = str(call_id_raw)
+                if tc_index is not None:
+                    existing_call_id = self._tool_call_id_by_index.get(tc_index)
+                    if existing_call_id is not None and existing_call_id != call_id:
+                        # Reusing the same index for multiple call_ids is ambiguous for id-less deltas.
+                        # Guard against silent misrouting by disabling index fallback for this index.
+                        self._ambiguous_tool_call_indexes.add(tc_index)
+                    self._tool_call_id_by_index[tc_index] = call_id
+            elif tc_index is not None:
+                if tc_index in self._ambiguous_tool_call_indexes:
+                    continue
+                mapped_call_id = self._tool_call_id_by_index.get(tc_index)
+                if mapped_call_id:
+                    call_id = mapped_call_id
+
+            if not call_id:
                 continue
-            call_id = str(call_id_raw)
 
             fn = tc.get("function") if isinstance(tc, dict) else getattr(tc, "function", None)
             fn_name = ""

schema.prisma

@@ -310,6 +310,16 @@ model LiteLLM_VerificationToken {
     litellm_budget_table LiteLLM_BudgetTable?   @relation(fields: [budget_id], references: [budget_id])
     litellm_organization_table LiteLLM_OrganizationTable?   @relation(fields: [organization_id], references: [organization_id])
     object_permission LiteLLM_ObjectPermissionTable?   @relation(fields: [object_permission_id], references: [object_permission_id])
+
+    // SELECT COUNT(*) FROM (SELECT "public"."LiteLLM_VerificationToken"."token" FROM "public"."LiteLLM_VerificationToken" WHERE ("public"."LiteLLM_VerificationToken"."user_id" = $1 AND ("public"."LiteLLM_VerificationToken"."team_id" IS NULL OR "public"."LiteLLM_VerificationToken"."team_id" <> $2)) OFFSET $3 ) AS "sub"
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE "public"."LiteLLM_VerificationToken"."user_id" = $1 OFFSET $2
+    @@index([user_id, team_id])
+
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE "public"."LiteLLM_VerificationToken"."team_id" = $1 OFFSET $2
+    @@index([team_id])
+
+    // SELECT ... FROM "public"."LiteLLM_VerificationToken" WHERE (("public"."LiteLLM_VerificationToken"."expires" IS NULL OR "public"."LiteLLM_VerificationToken"."expires" > $1) AND "public"."LiteLLM_VerificationToken"."budget_reset_at" < $2) OFFSET $3
+    @@index([budget_reset_at, expires])
 }
 
 // Audit table for deleted keys - preserves spend and key information for historical tracking

tests/mcp_tests/test_openapi_spec_path_url.py

@@ -0,0 +1,92 @@
+from __future__ import annotations
+
+from typing import Any, Dict
+
+import httpx
+import pytest
+
+from litellm.proxy._experimental.mcp_server import openapi_to_mcp_generator as gen
+
+
+class _FakeAsyncHTTPHandler:
+    """
+    Minimal stand-in for the object returned by get_async_httpx_client().
+    openapi_to_mcp_generator.load_openapi_spec_async() calls:
+
+        client = get_async_httpx_client(...)
+        r = await client.get(url, timeout=30.0)
+
+    So we must implement async get().
+    """
+
+    def __init__(self, response: httpx.Response, expected_url: str):
+        self._response = response
+        self._expected_url = expected_url
+        self.calls = 0
+
+    async def get(self, request_url: str, timeout: float = 30.0):
+        self.calls += 1
+        assert request_url == self._expected_url
+        assert timeout == 30.0
+        return self._response
+
+
+def test_load_openapi_spec_supports_http_url(monkeypatch: pytest.MonkeyPatch) -> None:
+    url = "http://example.local/openapi.json"
+    expected: Dict[str, Any] = {
+        "openapi": "3.0.0",
+        "info": {"title": "Test API", "version": "1.0.0"},
+        "paths": {},
+    }
+
+    # httpx.Response must include a Request for raise_for_status() to work.
+    req = httpx.Request("GET", url)
+    resp = httpx.Response(status_code=200, json=expected, request=req)
+
+    calls = {"get_async_httpx_client": 0}
+    handler_holder: Dict[str, Any] = {}
+
+    def fake_get_async_httpx_client(*args, **kwargs):
+        calls["get_async_httpx_client"] += 1
+        h = _FakeAsyncHTTPHandler(resp, expected_url=url)
+        handler_holder["handler"] = h
+        return h
+
+    # Ensure shared/custom client path is used
+    monkeypatch.setattr(gen, "get_async_httpx_client", fake_get_async_httpx_client)
+
+    # Fail loudly if someone reintroduces direct httpx.get()
+    def boom(*args, **kwargs):
+        raise AssertionError("Direct httpx.get() must not be used for URL spec loading")
+
+    monkeypatch.setattr(httpx, "get", boom)
+
+    spec = gen.load_openapi_spec(url)
+
+    assert spec == expected
+    assert calls["get_async_httpx_client"] == 1
+    assert handler_holder["handler"].calls == 1
+
+
+def test_load_openapi_spec_supports_local_file_path(tmp_path, monkeypatch: pytest.MonkeyPatch) -> None:
+    expected: Dict[str, Any] = {
+        "openapi": "3.0.0",
+        "info": {"title": "Local API", "version": "1.0.0"},
+        "paths": {},
+    }
+
+    p = tmp_path / "openapi.json"
+    p.write_text(
+        '{"openapi":"3.0.0","info":{"title":"Local API","version":"1.0.0"},"paths":{}}',
+        encoding="utf-8",
+    )
+
+    # For local files, shared client must NOT be used.
+    def boom_client(*args, **kwargs):
+        raise AssertionError("get_async_httpx_client() must not be called for local file paths")
+
+    monkeypatch.setattr(gen, "get_async_httpx_client", boom_client)
+
+    spec = gen.load_openapi_spec(str(p))
+    assert spec == expected
+