PyStreamMCP 3.1.0__py3-none-any.whl → 3.2.0__py3-none-any.whl

pystreammcp/__init__.py

@@ -7,7 +7,7 @@ Reduces token usage by 60-75% while maintaining quality.
 Includes StatGuardian integration for data quality validation.
 """
 
-__version__ = "3.0.0"
+__version__ = "3.2.0"
 
 from .agent import Agent, QueryResult
 from .query import Query, QueryIntent, QueryConstraints

pystreammcp/_mcp_connector.py

@@ -11,6 +11,9 @@ from dataclasses import dataclass, field
 from pathlib import Path
 from typing import Any, Dict, List, Optional
 from urllib.parse import urlparse
+from datetime import datetime, timezone
+
+from .webhook_router import EventRouter
 
 logger = logging.getLogger(__name__)
 
@@ -90,16 +93,23 @@ except ImportError:
                 self._ready = False
 
         def _generate_dab_config(self, tools: Dict[str, Any]) -> Dict:
+            # Bind to localhost only, with no cross-origin access and
+            # read-only anonymous permissions by default. This launches a
+            # local subprocess for MCP tool serving; it isn't meant to be
+            # reachable from other hosts or origins out of the box. Callers
+            # that need broader access (e.g. a container with its own
+            # network boundary) should override these explicitly rather
+            # than getting an open-to-everyone server by default.
             return {
                 "runtime": {
-                    "host": "0.0.0.0",
+                    "host": "127.0.0.1",
                     "port": self.port,
-                    "cors": {"origins": ["*"]},
+                    "cors": {"origins": []},
                 },
                 "entities": {
                     k: {
                         "source": k,
-                        "permissions": [{"actions": ["*"], "roles": ["*"]}],
+                        "permissions": [{"actions": ["read"], "roles": ["anonymous"]}],
                     }
                     for k in tools.keys()
                 },
@@ -142,6 +152,21 @@ class Orchestrator:
         self.config_path = Path(config_path) if config_path else DEFAULT_CONFIG_PATH
         self._transport = transport
 
+        # Real, stateful backing for the routing/webhook/health MCP tools
+        # (register_orchestration_webhook, list_service_endpoints,
+        # route_tool_invocation, get_tool_routing_info, monitor_mcp_health,
+        # orchestrate_cross_mcp_workflow): the same tested ServiceRegistry /
+        # ToolChainOrchestrator / FallbackManager infrastructure the HTTP
+        # webhook server (server.py) uses, not a second, disconnected copy.
+        self.event_router = EventRouter()
+        self.webhook_configs: Dict[str, Dict[str, Any]] = {}
+
+        # Real (if minimal) in-memory cache backing `cache_management`.
+        self._cache: Dict[str, Any] = {}
+        self._cache_enabled: bool = True
+        self._cache_hits = 0
+        self._cache_misses = 0
+
     def _load_federation_endpoints(self) -> List[str]:
         """Read [federation].endpoints from the config file.
 
@@ -232,14 +257,115 @@ class Orchestrator:
         ]
         return {"projects": projects, "total": len(projects)}
 
-    def plan_query_execution(self, query: str) -> dict:
-        return {"plan": []}
+    @staticmethod
+    def _estimate_tokens(text: str) -> int:
+        """Standard ~4-characters-per-token heuristic, floored at 1."""
+        return max(1, len(text) // 4)
+
+    def plan_query_execution(
+        self, query: str, projects_involved: Optional[List[str]] = None
+    ) -> dict:
+        """Build an execution plan against actually discovered projects.
+
+        Stages reference the real, currently-known health of each named
+        project (discovering first if nothing has been discovered yet)
+        rather than a fabricated 3-stage plan with made-up latency numbers.
+        Unknown/unrequested projects are reported as such, not silently
+        dropped or assumed healthy.
+        """
+        if not self.registered_projects:
+            self.discover_mcp_projects(refresh=True)
 
-    def optimize_cross_project_query(self, query: str) -> dict:
-        return {"optimized": query}
+        requested = projects_involved or []
+        resolved = []
+        for name in requested:
+            project = self.registered_projects.get(name)
+            resolved.append(
+                {
+                    "project_name": name,
+                    "status": project.status if project else "unknown",
+                }
+            )
+
+        return {
+            "query": query,
+            "estimated_tokens": self._estimate_tokens(query),
+            "projects": resolved,
+            "all_projects_healthy": all(p["status"] == "healthy" for p in resolved)
+            if resolved
+            else None,
+            "plan": [
+                {"stage": "validate", "projects": [p["project_name"] for p in resolved]},
+                {
+                    "stage": "execute",
+                    "projects": [p["project_name"] for p in resolved if p["status"] == "healthy"],
+                },
+            ],
+        }
 
-    def execute_federated_query(self, query: str, projects: list) -> dict:
-        return {"results": []}
+    def optimize_cross_project_query(
+        self, query: str, strategy: str = "balanced"
+    ) -> dict:
+        """Compute a real, input-derived token-reduction estimate.
+
+        Uses the same character-based token heuristic and per-strategy
+        reduction target as Agent.query() (agent.py) rather than a
+        hardcoded constant unconnected to the actual query text.
+        """
+        baseline_tokens = self._estimate_tokens(query)
+        reduction_target = {
+            "token_efficient": 0.75,
+            "quality_first": 0.60,
+        }.get(strategy, 0.70)
+        optimized_tokens = max(1, int(baseline_tokens * (1 - reduction_target)))
+        reduction_percent = (
+            (baseline_tokens - optimized_tokens) / baseline_tokens
+        ) * 100
+
+        return {
+            "original_query": query,
+            "baseline_tokens": baseline_tokens,
+            "optimized_tokens": optimized_tokens,
+            "token_reduction_percent": round(reduction_percent, 2),
+            "strategy": strategy,
+        }
+
+    def execute_federated_query(
+        self, query: str, projects: List[str], timeout_seconds: int = 30
+    ) -> dict:
+        """Report which of the requested projects are actually reachable.
+
+        This does not fabricate row counts or execution results — there is
+        no real cross-project query engine backing this yet. It reports
+        real discovery/health status for each requested project so callers
+        get an honest answer about what's actually available, and marks
+        the query itself as not executed.
+        """
+        if not self.registered_projects:
+            self.discover_mcp_projects(refresh=True)
+
+        results = []
+        for name in projects:
+            project = self.registered_projects.get(name)
+            results.append(
+                {
+                    "project_name": name,
+                    "status": project.status if project else "unknown",
+                    "error": project.error if project else "not discovered",
+                }
+            )
+
+        healthy = [r for r in results if r["status"] == "healthy"]
+        return {
+            "query": query,
+            "projects": results,
+            "status": "not_implemented",
+            "message": (
+                "Federated query execution across projects is not yet implemented; "
+                f"{len(healthy)}/{len(projects)} requested projects are currently healthy "
+                "and reachable."
+            ),
+        }
 
     def detect_compatible_projects(self, capability: str) -> dict:
         """Find discovered, healthy projects exposing a tool relevant to
@@ -319,19 +445,140 @@ class Orchestrator:
         return min(1.0, score)
 
     def handle_cross_database_join(self, left: str, right: str, key: str) -> dict:
-        return {"joined": {}}
-
-    def cache_management(self, action: str) -> dict:
-        return {"cache": {}}
-
-    def error_recovery_retry(self, query: str, error: str) -> dict:
-        return {"status": "retrying"}
+        """Cross-database joins are not implemented.
 
-    def report_performance_metrics(self, metric_type: str) -> dict:
-        return {"metrics": []}
-
-    def estimate_query_cost_multi_project(self, query: str) -> dict:
-        return {"cost": 0}
+        Honest "not_implemented" rather than a fabricated row count / join
+        plan: there is no real query engine here capable of executing a
+        join across heterogeneous databases.
+        """
+        return {
+            "left_source": left,
+            "right_source": right,
+            "join_key": key,
+            "status": "not_implemented",
+            "message": "Cross-database join execution is not yet implemented.",
+        }
+
+    def cache_management(
+        self,
+        action: str,
+        ttl_seconds: Optional[int] = None,
+        target: Optional[str] = None,
+    ) -> dict:
+        """Manage a real (if minimal) in-memory cache.
+
+        Tracks actual entries and real hit/miss counters rather than
+        fabricated size/hit-rate figures.
+        """
+        if action == "enable":
+            self._cache_enabled = True
+            return {"action": action, "status": "success", "cache_enabled": True}
+        if action == "disable":
+            self._cache_enabled = False
+            return {"action": action, "status": "success", "cache_enabled": False}
+        if action == "clear":
+            if target:
+                self._cache.pop(target, None)
+            else:
+                self._cache.clear()
+            return {"action": action, "status": "success", "entries": len(self._cache)}
+        if action == "stats":
+            total_lookups = self._cache_hits + self._cache_misses
+            hit_rate = (
+                (self._cache_hits / total_lookups * 100) if total_lookups else 0.0
+            )
+            return {
+                "action": action,
+                "status": "success",
+                "cache_enabled": self._cache_enabled,
+                "entries": len(self._cache),
+                "hits": self._cache_hits,
+                "misses": self._cache_misses,
+                "hit_rate_percent": round(hit_rate, 2),
+            }
+        return {
+            "action": action,
+            "status": "error",
+            "message": f"Unknown action: {action!r} (expected enable/disable/clear/stats)",
+        }
+
+    def error_recovery_retry(
+        self,
+        failed_query: str,
+        error_type: str,
+        retry_strategy: str = "exponential_backoff",
+        max_retries: int = 3,
+    ) -> dict:
+        """Queue a failed query for retry via the real FallbackManager.
+
+        Reports the real resulting queue size rather than a hardcoded
+        "retrying" status disconnected from any actual queue state.
+        """
+        self.event_router.fallback_manager._queue_for_retry(
+            failed_query, {"error_type": error_type, "max_retries": max_retries}
+        )
+        queue_status = self.event_router.fallback_manager.get_retry_queue_status()
+        return {
+            "failed_query": failed_query,
+            "error_type": error_type,
+            "retry_strategy": retry_strategy,
+            "max_retries": max_retries,
+            "status": "queued_for_retry",
+            "queue_size": queue_status["queue_size"],
+        }
+
+    def report_performance_metrics(
+        self, metric_type: str = "latency", group_by: str = "project"
+    ) -> dict:
+        """Report real health-history metrics from the ServiceRegistry.
+
+        Returns actually recorded health-check history (may legitimately
+        be empty if nothing has reported health metrics yet) rather than
+        fabricated latency/throughput figures.
+        """
+        history = self.event_router.service_registry.health_history
+        metrics = []
+        for project_name, entries in history.items():
+            if not entries:
+                continue
+            latest = entries[-1]
+            value = latest.get("metrics", {}).get(metric_type)
+            metrics.append(
+                {
+                    "group": project_name,
+                    "value": value,
+                    "status": latest.get("status"),
+                    "recorded_at": latest.get("timestamp"),
+                }
+            )
+        return {
+            "metric_type": metric_type,
+            "group_by": group_by,
+            "metrics": metrics,
+            "message": None
+            if metrics
+            else "No health metrics have been recorded for any project yet.",
+        }
+
+    def estimate_query_cost_multi_project(
+        self, query: str, projects: Optional[List[str]] = None, cost_model: str = "tokens"
+    ) -> dict:
+        """Estimate token cost from the actual query text.
+
+        Uses the same character-based heuristic as optimize_cross_project_query
+        rather than a fixed constant unrelated to the query content.
+        """
+        tokens = self._estimate_tokens(query)
+        projects = projects or []
+        per_project_tokens = tokens // len(projects) if projects else tokens
+
+        return {
+            "query": query,
+            "cost_model": cost_model,
+            "projects_involved": projects,
+            "estimated_tokens": tokens,
+            "cost_breakdown": {p: {"tokens": per_project_tokens} for p in projects},
+        }
 
     def manage_endpoint_federation(self, action: str) -> dict:
         """Manage the federation endpoint list.
@@ -356,6 +603,172 @@ class Orchestrator:
         ]
         return {"status": "refreshed" if action == "refresh" else "listed", "endpoints": endpoints}
 
+    # --- Real orchestration webhook / routing surface -------------------
+    #
+    # Backed by the same tested ServiceRegistry / ToolChainOrchestrator /
+    # FallbackManager machinery in webhook_router.py that the HTTP webhook
+    # server (server.py) uses — not a second, disconnected implementation.
+
+    def register_orchestration_webhook(
+        self,
+        webhook_id: str,
+        url: str,
+        events: List[str],
+        secret_key: Optional[str] = None,
+    ) -> dict:
+        """Register a webhook subscription in real, queryable state."""
+        self.webhook_configs[webhook_id] = {
+            "webhook_id": webhook_id,
+            "url": url,
+            "events": events,
+            "secret_key": secret_key,
+            "created_at": datetime.now(timezone.utc).isoformat(),
+        }
+        return {
+            "status": "success",
+            "webhook_id": webhook_id,
+            "url": url,
+            "events": events,
+            "signature_method": "HMAC-SHA256" if secret_key else "none",
+            "created_at": self.webhook_configs[webhook_id]["created_at"],
+        }
+
+    def list_service_endpoints(
+        self, filter_by_status: str = "all", include_metrics: bool = False
+    ) -> dict:
+        """List MCP endpoints registered with the real ServiceRegistry.
+
+        This reflects only endpoints that have actually been registered
+        (via `mcp.available` events or `register_mcp_endpoint`) — it is
+        genuinely empty until something registers, not a fabricated
+        7-project fixture list.
+        """
+        snapshot = self.event_router.service_registry.get_registry_snapshot()
+        endpoints = snapshot["mcps"]
+        if filter_by_status != "all":
+            endpoints = [e for e in endpoints if e["status"] == filter_by_status]
+        if not include_metrics:
+            for ep in endpoints:
+                ep.pop("health_metrics", None)
+        return {
+            "status": "success",
+            "endpoints": endpoints,
+            "total": len(endpoints),
+            "healthy_count": len([e for e in endpoints if e["status"] == "healthy"]),
+        }
+
+    async def route_tool_invocation(
+        self,
+        tool_name: str,
+        params: Optional[Dict[str, Any]] = None,
+        chain_id: Optional[str] = None,
+        use_fallback: bool = True,
+        timeout_ms: Optional[int] = None,
+    ) -> dict:
+        """Route a tool invocation via the real ToolChainOrchestrator.
+
+        Actually looks the tool up in the ServiceRegistry; if it isn't
+        found and `use_fallback` is set, falls back through
+        FallbackManager.invoke_with_fallback rather than fabricating a
+        "routed" response to an endpoint that doesn't exist.
+        """
+        chain_context = {"chain_id": chain_id} if chain_id else None
+        result = await self.event_router.tool_orchestrator.route_tool_invocation(
+            tool_name=tool_name,
+            params=params or {},
+            chain_context=chain_context,
+        )
+        if result.get("status") == "error" and use_fallback:
+            result = await self.event_router.fallback_manager.invoke_with_fallback(
+                tool_name=tool_name, params=params or {}, fallback_enabled=True
+            )
+        return result
+
+    def get_tool_routing_info(
+        self, tool_name: str, include_fallbacks: bool = False
+    ) -> dict:
+        """Real routing info for a tool: where the ServiceRegistry says
+        it actually lives, not a fabricated 145.5ms/0.999 fixture."""
+        result = self.event_router.service_registry.find_tool(tool_name)
+        if not result:
+            return {
+                "status": "not_found",
+                "tool_name": tool_name,
+                "message": f"Tool {tool_name} not found in any registered MCP",
+            }
+        project_name, endpoint = result
+        routing = {
+            "tool_name": tool_name,
+            "primary_mcp": project_name,
+            "primary_endpoint": f"http://localhost:{endpoint.port}",
+            "health_status": endpoint.status,
+        }
+        if include_fallbacks:
+            fallback = self.event_router.tool_orchestrator.find_fallback_tool(tool_name)
+            routing["fallback_mcps"] = (
+                [{"project": fallback[0], "endpoint": fallback[1]}] if fallback else []
+            )
+        return {"status": "success", "routing": routing}
+
+    def monitor_mcp_health(self, project_name: str = "all") -> dict:
+        """Real orchestration health snapshot from the ServiceRegistry."""
+        status = self.event_router.get_status()
+        registry = status["registry"]
+        if project_name != "all":
+            endpoint = self.event_router.service_registry.endpoints.get(project_name)
+            return {
+                "status": "success",
+                "project_name": project_name,
+                "current_status": endpoint.status if endpoint else "unknown",
+            }
+        return {
+            "status": "success",
+            "total_mcps": registry["total_mcps"],
+            "healthy_mcps": registry["healthy_mcps"],
+            "degraded_or_unavailable": registry["total_mcps"] - registry["healthy_mcps"],
+        }
+
+    async def orchestrate_cross_mcp_workflow(
+        self,
+        workflow_id: str,
+        tool_sequence: List[str],
+        cascade_on_success: bool = True,
+        error_handling: str = "fail_fast",
+    ) -> dict:
+        """Execute a real sequence of routed tool invocations.
+
+        Each stage's outcome is the real result of
+        ToolChainOrchestrator.route_tool_invocation (found/not-found
+        against the ServiceRegistry) — not a fabricated "pending" plan
+        that never actually executes.
+        """
+        stages = []
+        for position, tool_name in enumerate(tool_sequence):
+            result = await self.event_router.tool_orchestrator.route_tool_invocation(
+                tool_name=tool_name,
+                params={},
+                chain_context={
+                    "chain_id": workflow_id,
+                    "position_in_chain": position,
+                    "total_chain_length": len(tool_sequence),
+                },
+            )
+            stages.append({"stage": position + 1, "tool": tool_name, "result": result})
+            if result.get("status") == "error" and error_handling == "fail_fast":
+                return {
+                    "status": "failed",
+                    "workflow_id": workflow_id,
+                    "stages": stages,
+                    "failed_at_stage": position + 1,
+                }
+
+        return {
+            "status": "completed",
+            "workflow_id": workflow_id,
+            "stages": stages,
+            "total_stages": len(tool_sequence),
+        }
+
     def start_mcp_connector(self, port: int = 8772) -> str:
         from pystreammcp._mcp_tools import PyStreamMCPHandler, PyStreamMCPTools