feat: Add 6 new OpenRouter models (April 2026 refresh)

config.py

@@ -14,9 +14,9 @@
 # These values are used in server responses and for tracking releases
 # IMPORTANT: This is the single source of truth for version and author info
 # Semantic versioning: MAJOR.MINOR.PATCH
-__version__ = "9.8.2"
+__version__ = "9.9.0"
 # Last update date in ISO format
-__updated__ = "2025-12-15"
+__updated__ = "2026-03-06"
 # Primary maintainer
 __author__ = "Fahad Gilani"
 

docs/custom_models.md

@@ -52,18 +52,26 @@ The curated defaults in `conf/openrouter_models.json` include popular entries su
 
 | Alias | Canonical Model | Highlights |
 |-------|-----------------|------------|
-| `opus`, `claude-opus` | `anthropic/claude-opus-4.1` | Flagship Claude reasoning model with vision |
-| `sonnet`, `sonnet4.5` | `anthropic/claude-sonnet-4.5` | Balanced Claude with high context window |
+| `opus`, `claude-opus` | `anthropic/claude-opus-4.6` | Latest Anthropic flagship (1M context, vision). `opus4.5` → 4.5, `opus4.1` → 4.1 |
+| `sonnet` | `anthropic/claude-sonnet-4.6` | Frontier Sonnet (1M context, vision). `sonnet4.5` → 4.5 |
 | `haiku` | `anthropic/claude-3.5-haiku` | Fast Claude option with vision |
-| `pro`, `gemini` | `google/gemini-2.5-pro` | Frontier Gemini with extended thinking |
+| `pro`, `gemini` | `google/gemini-3.1-pro-preview` | Latest Gemini Pro with 1M context, thinking. `gemini3.0` → 3.0 |
 | `flash` | `google/gemini-2.5-flash` | Ultra-fast Gemini with vision |
-| `mistral` | `mistralai/mistral-large-2411` | Frontier Mistral (text only) |
-| `llama3` | `meta-llama/llama-3-70b` | Large open-weight text model |
-| `deepseek-r1` | `deepseek/deepseek-r1-0528` | DeepSeek reasoning model |
-| `perplexity` | `perplexity/llama-3-sonar-large-32k-online` | Search-augmented model |
+| `gpt5`, `gpt5.4` | `openai/gpt-5.4` | Unified frontier model (1M context, 128K output). `gpt5.0` → 5.0 |
+| `gpt5pro` | `openai/gpt-5.4-pro` | Enhanced reasoning variant (1M context). `gpt5.2-pro` → 5.2 Pro |
+| `codex`, `codex-5.3` | `openai/gpt-5.3-codex` | Latest agentic coding model (Responses API). `codex-5.0` → 5.0 |
 | `gpt5.2`, `gpt-5.2`, `5.2` | `openai/gpt-5.2` | Flagship GPT-5.2 with reasoning and vision |
 | `gpt5.1-codex`, `codex-5.1` | `openai/gpt-5.1-codex` | Agentic coding specialization (Responses API) |
-| `codex-mini`, `gpt5.1-codex-mini` | `openai/gpt-5.1-codex-mini` | Cost-efficient Codex variant with streaming |
+| `codex-mini` | `openai/gpt-5.1-codex-mini` | Cost-efficient Codex variant with streaming |
+| `mistral` | `mistralai/mistral-large-2411` | Frontier Mistral (text only) |
+| `devstral` | `mistralai/devstral-2512` | 123B agentic coding model (262K context) |
+| `deepseek-r1` | `deepseek/deepseek-r1-0528` | DeepSeek reasoning model |
+| `deepseek-v3`, `dsv3` | `deepseek/deepseek-v3.2-exp` | DeepSeek V3.2 with strong reasoning (164K context) |
+| `qwen` | `qwen/qwen3.6-plus` | Alibaba's latest frontier model (1M context). `qwen3.5` → 3.5 |
+| `minimax` | `minimax/minimax-m2.7` | 1T param model matching GPT-5.3-Codex. `m2.5` → 2.5 |
+| `hermes` | `nousresearch/hermes-4-405b` | Hybrid reasoning with deliberative alignment (131K context) |
+| `qwen-coder` | `qwen/qwen3-coder` | Frontier open-weight coding model, 480B MoE (262K context) |
+| `flash-lite` | `google/gemini-3.1-flash-lite-preview` | Cost-efficient Gemini flash variant (1M context) |
 
 Consult the JSON file for the full list, aliases, and capability flags. Add new entries as OpenRouter releases additional models.
 

pyproject.toml

@@ -1,8 +1,8 @@
 [project]
 name = "pal-mcp-server"
-version = "9.8.2"
+version = "9.9.0"
 description = "AI-powered MCP server with multiple model providers"
-requires-python = ">=3.9"
+requires-python = ">=3.10"
 dependencies = [
     "mcp>=1.0.0",
     "google-genai>=1.19.0",

server.py

@@ -689,6 +689,66 @@ async def handle_list_tools() -> list[Tool]:
     return tools
 
 
+async def _execute_with_progress(coro):
+    """
+    Run a tool coroutine while sending MCP progress notifications every ~15 seconds.
+
+    This prevents Claude Desktop's ~4 minute timeout from killing long-running
+    tool calls (e.g. slow model API responses that take 4-8 minutes).
+    """
+    # Get progress token from request context
+    progress_token = None
+    session = None
+    request_id = None
+    try:
+        ctx = server.request_context
+        if ctx.meta:
+            progress_token = ctx.meta.progressToken
+        session = ctx.session
+        request_id = ctx.request_id
+    except (LookupError, AttributeError):
+        pass
+
+    if session is None:
+        return await coro
+
+    # Claude Desktop omits progressToken but still times out after ~4 min.
+    # Use request_id as fallback — sending any notification keeps stdio alive.
+    if progress_token is None:
+        progress_token = request_id if request_id is not None else "pal-progress"
+
+    task = asyncio.create_task(coro)
+    tick = 0
+    notification_failed = False
+    try:
+        while not task.done():
+            try:
+                await asyncio.wait_for(asyncio.shield(task), timeout=15.0)
+            except asyncio.TimeoutError:
+                tick += 1
+                elapsed = tick * 15
+                try:
+                    await session.send_progress_notification(
+                        progress_token=progress_token,
+                        progress=float(tick),
+                        message=f"Model is thinking... ({elapsed}s elapsed)",
+                        related_request_id=request_id,
+                    )
+                except Exception as exc:
+                    if not notification_failed:
+                        logger.warning(f"Progress notification failed: {exc}")
+                        notification_failed = True
+    except asyncio.CancelledError:
+        task.cancel()
+        try:
+            await task
+        except Exception as exc:
+            logger.debug(f"Task raised during cancellation teardown: {exc}")
+        raise
+
+    return task.result()
+
+
 @server.call_tool()
 async def handle_call_tool(name: str, arguments: dict[str, Any]) -> list[TextContent]:
     """
@@ -807,7 +867,7 @@ async def handle_call_tool(name: str, arguments: dict[str, Any]) -> list[TextCon
         if not tool.requires_model():
             logger.debug(f"Tool {name} doesn't require model resolution - skipping model validation")
             # Execute tool directly without model context
-            return await tool.execute(arguments)
+            return await _execute_with_progress(tool.execute(arguments))
 
         # Handle auto mode at MCP boundary - resolve to specific model
         if model_name.lower() == "auto":
@@ -862,7 +922,7 @@ async def handle_call_tool(name: str, arguments: dict[str, Any]) -> list[TextCon
                 raise ToolExecutionError(ToolOutput(**file_size_check).model_dump_json())
 
         # Execute tool with pre-resolved model context
-        result = await tool.execute(arguments)
+        result = await _execute_with_progress(tool.execute(arguments))
         logger.info(f"Tool '{name}' execution completed")
 
         # Log completion to activity file

simulator_tests/test_chat_simple_validation.py

@@ -13,7 +13,6 @@
 - Conversation context preservation across turns
 """
 
-
 from .conversation_base_test import ConversationBaseTest
 
 

simulator_tests/test_conversation_chain_validation.py

@@ -21,7 +21,6 @@
 - Properly traverse parent relationships for history reconstruction
 """
 
-
 from .conversation_base_test import ConversationBaseTest
 
 

simulator_tests/test_cross_tool_comprehensive.py

@@ -12,7 +12,6 @@
 5. Proper tool chaining with context
 """
 
-
 from .conversation_base_test import ConversationBaseTest
 
 

simulator_tests/test_ollama_custom_url.py

@@ -9,7 +9,6 @@
 - Model alias resolution for local models
 """
 
-
 from .base_test import BaseSimulatorTest
 
 

simulator_tests/test_openrouter_fallback.py

@@ -8,7 +8,6 @@
 - Auto mode correctly selects OpenRouter models
 """
 
-
 from .base_test import BaseSimulatorTest
 
 

simulator_tests/test_openrouter_models.py

@@ -9,7 +9,6 @@
 - Error handling when models are not available
 """
 
-
 from .base_test import BaseSimulatorTest
 
 

simulator_tests/test_xai_models.py

@@ -9,7 +9,6 @@
 - API integration and response validation
 """
 
-
 from .base_test import BaseSimulatorTest
 
 

tests/test_custom_provider.py

@@ -87,9 +87,9 @@ def test_model_alias_resolution(self):
         provider = CustomProvider(api_key="test-key", base_url="http://localhost:11434/v1")
 
         # Test that aliases resolve properly
-        # "llama" now resolves to "meta-llama/llama-3-70b" (the OpenRouter model)
-        resolved = provider._resolve_model_name("llama")
-        assert resolved == "meta-llama/llama-3-70b"
+        # "deepseek" resolves to "deepseek/deepseek-r1-0528" (the OpenRouter model)
+        resolved = provider._resolve_model_name("deepseek")
+        assert resolved == "deepseek/deepseek-r1-0528"
 
         # Test local model alias
         resolved_local = provider._resolve_model_name("local-llama")
@@ -216,8 +216,8 @@ def custom_provider_factory(api_key=None):
             custom_provider = custom_provider_factory()
             openrouter_provider = OpenRouterProvider(api_key="test-openrouter-key")
 
-            assert not custom_provider.validate_model_name("llama")
-            assert openrouter_provider.validate_model_name("llama")
+            assert not custom_provider.validate_model_name("deepseek")
+            assert openrouter_provider.validate_model_name("deepseek")
 
 
 class TestConfigureProvidersFunction:

tests/test_directory_expansion_tracking.py

@@ -37,27 +37,23 @@ def temp_directory_with_files(self, project_path):
         files = []
         for i in range(5):
             swift_file = temp_path / f"File{i}.swift"
-            swift_file.write_text(
-                f"""
+            swift_file.write_text(f"""
 import Foundation
 
 class TestClass{i} {{
     func testMethod{i}() -> String {{
         return "test{i}"
     }}
 }}
-"""
-            )
+""")
             files.append(str(swift_file))
 
         # Create a Python file as well
         python_file = temp_path / "helper.py"
-        python_file.write_text(
-            """
+        python_file.write_text("""
 def helper_function():
     return "helper"
-"""
-        )
+""")
         files.append(str(python_file))
 
         try:

tests/test_docker_implementation.py

@@ -310,13 +310,11 @@ def temp_project_dir():
 
         # Create base files
         (temp_path / "server.py").write_text("# Mock server.py")
-        (temp_path / "Dockerfile").write_text(
-            """
+        (temp_path / "Dockerfile").write_text("""
 FROM python:3.11-slim
 COPY server.py /app/
 CMD ["python", "/app/server.py"]
-"""
-        )
+""")
 
         yield temp_path
 

tests/test_model_resolution_bug.py

@@ -22,21 +22,27 @@ def setup_method(self):
         self.consensus_tool = ConsensusTool()
 
     def test_openrouter_registry_resolves_gemini_alias(self):
-        """Test that OpenRouter registry properly resolves 'gemini' to 'google/gemini-3-pro-preview'."""
+        """Test that OpenRouter registry properly resolves 'gemini' to the latest Gemini Pro."""
         # Test the registry directly
         provider = OpenRouterProvider("test_key")
 
-        # Test alias resolution
+        # Test alias resolution - 'gemini' and 'pro' now point to 3.1
         resolved_model_name = provider._resolve_model_name("gemini")
         assert (
-            resolved_model_name == "google/gemini-3-pro-preview"
-        ), f"Expected 'google/gemini-3-pro-preview', got '{resolved_model_name}'"
+            resolved_model_name == "google/gemini-3.1-pro-preview"
+        ), f"Expected 'google/gemini-3.1-pro-preview', got '{resolved_model_name}'"
 
         # Test that it also works with 'pro' alias
         resolved_pro = provider._resolve_model_name("pro")
         assert (
-            resolved_pro == "google/gemini-3-pro-preview"
-        ), f"Expected 'google/gemini-3-pro-preview', got '{resolved_pro}'"
+            resolved_pro == "google/gemini-3.1-pro-preview"
+        ), f"Expected 'google/gemini-3.1-pro-preview', got '{resolved_pro}'"
+
+        # Test version-specific alias still works for backward compat
+        resolved_3_0 = provider._resolve_model_name("gemini3.0")
+        assert (
+            resolved_3_0 == "google/gemini-3-pro-preview"
+        ), f"Expected 'google/gemini-3-pro-preview', got '{resolved_3_0}'"
 
     # DELETED: test_provider_registry_returns_openrouter_for_gemini
     # This test had a flawed mock setup - it mocked get_provider() but called get_provider_for_model().

tests/test_openrouter_provider.py

@@ -78,28 +78,32 @@ def test_model_alias_resolution(self):
         """Test model alias resolution."""
         provider = OpenRouterProvider(api_key="test-key")
 
-        # Test alias resolution
-        assert provider._resolve_model_name("opus") == "anthropic/claude-opus-4.5"
+        # Test alias resolution - generic aliases point to latest versions
+        assert provider._resolve_model_name("opus") == "anthropic/claude-opus-4.6"
+        assert provider._resolve_model_name("opus4.6") == "anthropic/claude-opus-4.6"
         assert provider._resolve_model_name("opus4.5") == "anthropic/claude-opus-4.5"
         assert provider._resolve_model_name("opus4.1") == "anthropic/claude-opus-4.1"
-        assert provider._resolve_model_name("sonnet") == "anthropic/claude-sonnet-4.5"
+        assert provider._resolve_model_name("sonnet") == "anthropic/claude-sonnet-4.6"
+        assert provider._resolve_model_name("sonnet4.6") == "anthropic/claude-sonnet-4.6"
+        assert provider._resolve_model_name("sonnet4.5") == "anthropic/claude-sonnet-4.5"
         assert provider._resolve_model_name("sonnet4.1") == "anthropic/claude-sonnet-4.1"
         assert provider._resolve_model_name("o3") == "openai/o3"
         assert provider._resolve_model_name("o3-mini") == "openai/o3-mini"
         assert provider._resolve_model_name("o3mini") == "openai/o3-mini"
         assert provider._resolve_model_name("o4-mini") == "openai/o4-mini"
-        assert provider._resolve_model_name("o4-mini") == "openai/o4-mini"
         assert provider._resolve_model_name("haiku") == "anthropic/claude-3.5-haiku"
         assert provider._resolve_model_name("mistral") == "mistralai/mistral-large-2411"
         assert provider._resolve_model_name("grok-4") == "x-ai/grok-4"
         assert provider._resolve_model_name("grok4") == "x-ai/grok-4"
         assert provider._resolve_model_name("grok") == "x-ai/grok-4"
         assert provider._resolve_model_name("deepseek") == "deepseek/deepseek-r1-0528"
         assert provider._resolve_model_name("r1") == "deepseek/deepseek-r1-0528"
+        assert provider._resolve_model_name("pro") == "google/gemini-3.1-pro-preview"
+        assert provider._resolve_model_name("gemini3.0") == "google/gemini-3-pro-preview"
 
         # Test case-insensitive
-        assert provider._resolve_model_name("OPUS") == "anthropic/claude-opus-4.5"
-        assert provider._resolve_model_name("SONNET") == "anthropic/claude-sonnet-4.5"
+        assert provider._resolve_model_name("OPUS") == "anthropic/claude-opus-4.6"
+        assert provider._resolve_model_name("SONNET") == "anthropic/claude-sonnet-4.6"
         assert provider._resolve_model_name("O3") == "openai/o3"
         assert provider._resolve_model_name("Mistral") == "mistralai/mistral-large-2411"
 
@@ -307,18 +311,18 @@ def test_registry_capabilities(self):
 
         registry = OpenRouterModelRegistry()
 
-        # Test known model (opus alias now points to 4.5)
+        # Test known model (opus alias now points to 4.6)
         caps = registry.get_capabilities("opus")
         assert caps is not None
-        assert caps.model_name == "anthropic/claude-opus-4.5"
-        assert caps.context_window == 200000  # Claude's context window
+        assert caps.model_name == "anthropic/claude-opus-4.6"
+        assert caps.context_window == 1000000  # Claude 4.6 context window
 
         # Test using full model name for 4.5
         caps = registry.get_capabilities("anthropic/claude-opus-4.5")
         assert caps is not None
         assert caps.model_name == "anthropic/claude-opus-4.5"
 
-        # Test opus4.5 alias
+        # Test opus4.5 alias (backward compat)
         caps = registry.get_capabilities("opus4.5")
         assert caps is not None
         assert caps.model_name == "anthropic/claude-opus-4.5"
@@ -343,12 +347,14 @@ def test_multiple_aliases_same_model(self):
 
         registry = OpenRouterModelRegistry()
 
-        # All these should resolve to Claude Sonnet 4.5
-        sonnet_45_aliases = ["sonnet", "sonnet4.5"]
-        for alias in sonnet_45_aliases:
-            config = registry.resolve(alias)
-            assert config is not None
-            assert config.model_name == "anthropic/claude-sonnet-4.5"
+        # 'sonnet' now resolves to 4.6, 'sonnet4.5' still resolves to 4.5
+        config = registry.resolve("sonnet")
+        assert config is not None
+        assert config.model_name == "anthropic/claude-sonnet-4.6"
+
+        config = registry.resolve("sonnet4.5")
+        assert config is not None
+        assert config.model_name == "anthropic/claude-sonnet-4.5"
 
         # Test Sonnet 4.1 alias
         config = registry.resolve("sonnet4.1")