Feat/add insforge + supabase mcp support to evaluate against postgres

run-task.sh

@@ -139,6 +139,16 @@ elif [ "$SERVICE" = "filesystem" ]; then
         $([ -f .mcp_env ] && echo "-v $(pwd)/.mcp_env:/app/.mcp_env:ro") \
         "$DOCKER_IMAGE" \
         python3 -m pipeline --mcp "$SERVICE" --k 1 "$@"
+elif [ "$SERVICE" = "insforge" ]; then
+    # For Insforge service, use host network to access Insforge backend on host
+    docker run --rm \
+        --memory="$DOCKER_MEMORY_LIMIT" \
+        --cpus="$DOCKER_CPU_LIMIT" \
+        --add-host=host.docker.internal:host-gateway \
+        -v "$(pwd)/results:/app/results" \
+        $([ -f .mcp_env ] && echo "-v $(pwd)/.mcp_env:/app/.mcp_env:ro") \
+        "$DOCKER_IMAGE" \
+        python3 -m pipeline --mcp "$SERVICE" --k 1 "$@"
 else
     # For other services (notion, github, playwright, etc.)
     docker run --rm \

src/agents/base_agent.py

@@ -19,8 +19,8 @@
 class BaseMCPAgent(ABC):
     """Base class with shared functionality for MCPMark agents."""
 
-    STDIO_SERVICES = ["notion", "filesystem", "playwright", "playwright_webarena", "postgres"]
-    HTTP_SERVICES = ["github"]
+    STDIO_SERVICES = ["notion", "filesystem", "playwright", "playwright_webarena", "postgres", "insforge"]
+    HTTP_SERVICES = ["github", "supabase"]
     DEFAULT_TIMEOUT = 600
 
     CLAUDE_THINKING_BUDGETS = {
@@ -207,6 +207,20 @@ def _create_stdio_server(self) -> MCPStdioServer:
                 env={"DATABASE_URI": database_url},
             )
 
+        if self.mcp_service == "insforge":
+            api_key = self.service_config.get("api_key")
+            backend_url = self.service_config.get("backend_url")
+            if not all([api_key, backend_url]):
+                raise ValueError("Insforge requires api_key and backend_url")
+            return MCPStdioServer(
+                command="npx",
+                args=["-y", "@insforge/mcp@dev"],
+                env={
+                    "INSFORGE_API_KEY": api_key,
+                    "INSFORGE_BACKEND_URL": backend_url,
+                },
+            )
+
         raise ValueError(f"Unsupported stdio service: {self.mcp_service}")
 
     def _create_http_server(self) -> MCPHttpServer:

src/agents/mcpmark_agent.py

@@ -844,18 +844,32 @@ def _create_stdio_server(self) -> MCPStdioServer:
             username = self.service_config.get("username")
             password = self.service_config.get("password")
             database = self.service_config.get("current_database") or self.service_config.get("database")
-            
+
             if not all([username, password, database]):
                 raise ValueError("PostgreSQL requires username, password, and database")
-            
+
             database_url = f"postgresql://{username}:{password}@{host}:{port}/{database}"
-            
+
             return MCPStdioServer(
                 command="pipx",
                 args=["run", "postgres-mcp", "--access-mode=unrestricted"],
                 env={"DATABASE_URI": database_url}
             )
-
+
+        elif self.mcp_service == "insforge":
+            api_key = self.service_config.get("api_key")
+            backend_url = self.service_config.get("backend_url")
+            if not all([api_key, backend_url]):
+                raise ValueError("Insforge requires api_key and backend_url")
+            return MCPStdioServer(
+                command="npx",
+                args=["-y", "@insforge/mcp@dev"],
+                env={
+                    "INSFORGE_API_KEY": api_key,
+                    "INSFORGE_BACKEND_URL": backend_url,
+                },
+            )
+
         else:
             raise ValueError(f"Unsupported stdio service: {self.mcp_service}")
 
@@ -866,14 +880,34 @@ def _create_http_server(self) -> MCPHttpServer:
             github_token = self.service_config.get("github_token")
             if not github_token:
                 raise ValueError("GitHub token required")
-            
+
             return MCPHttpServer(
                 url="https://api.githubcopilot.com/mcp/",
                 headers={
                     "Authorization": f"Bearer {github_token}",
                     "User-Agent": "MCPMark/1.0"
                 }
             )
+
+        elif self.mcp_service == "supabase":
+            # Use built-in MCP server from Supabase CLI
+            api_url = self.service_config.get("api_url", "http://localhost:54321")
+            api_key = self.service_config.get("api_key", "")
+
+            if not api_key:
+                raise ValueError("Supabase requires api_key (use secret key from 'supabase status')")
+
+            # Supabase CLI exposes MCP at /mcp endpoint
+            mcp_url = f"{api_url}/mcp"
+
+            return MCPHttpServer(
+                url=mcp_url,
+                headers={
+                    "apikey": api_key,
+                    "Authorization": f"Bearer {api_key}",
+                }
+            )
+
         else:
             raise ValueError(f"Unsupported HTTP service: {self.mcp_service}")
 

src/aggregators/aggregate_results.py

@@ -23,19 +23,19 @@
 def discover_tasks() -> Dict[str, List[str]]:
     """Discover all tasks from ./tasks directory."""
     tasks_dir = Path("./tasks")
-    
+
     all_tasks = {}
-    
+
     # Handle each MCP service
     # Note: playwright and playwright_webarena both map to "playwright" MCP
     service_mappings = {
         "filesystem": ["filesystem"],
         "github": ["github"],
         "notion": ["notion"],
         "playwright": ["playwright", "playwright_webarena"],  # Both count as playwright
-        "postgres": ["postgres"]
+        "postgres": ["postgres", "supabase", "insforge"]  # All map to postgres
     }
-    
+
     for mcp_service, task_dirs in service_mappings.items():
         tasks = []
         for task_dir_name in task_dirs:
@@ -68,9 +68,11 @@ def collect_results(exp_dir: Path, k: int) -> Dict[str, Dict[str, Any]]:
             continue
 
         model, service = model_service_dir.name.split("__", 1)
-        # Normalize service name: treat playwright_webarena as playwright
+        # Normalize service names
         if service == "playwright_webarena":
             service = "playwright"
+        elif service in ["supabase", "insforge"]:
+            service = "postgres"
 
         for run_idx in range(1, k + 1):
             run_dir = model_service_dir / f"run-{run_idx}"
@@ -874,23 +876,23 @@ def main():
         help="Comma-separated list of models that only need run-1"
     )
     parser.add_argument("--push", action="store_true", help="Push to GitHub (default to main)")
-    
+
     args = parser.parse_args()
-    
+
     # Parse single-run models
     single_run_models = []
     if args.single_run_models:
         single_run_models = [m.strip() for m in args.single_run_models.split(",")]
         print(f"📌 Single-run models: {', '.join(single_run_models)}")
-    
+
     # Setup paths
     exp_dir = Path("./results") / args.exp_name
     if not exp_dir.exists():
         print(f"❌ Experiment directory {exp_dir} does not exist")
         return 1
-    
+
     print(f"🔄 Processing experiment: {args.exp_name}")
-    
+
     # Discover all tasks
     print("📋 Discovering tasks...")
     all_tasks = discover_tasks()