More visibility for gateway and tool status and auto healing capabilities (#215)

* changes implemented reachable and enabled as parameters for status of gateways and tools

Signed-off-by: <Keval Mahajan> <[email protected]>

* Updated test cases as per new changes for status in gateways and tools

Signed-off-by: <Keval Mahajan> <[email protected]>

* minor changes

Signed-off-by: <Keval Mahajan> <[email protected]>

* black and isort

Signed-off-by: <Keval Mahajan> <[email protected]>

* Alembic script for db migration

Signed-off-by: <Keval Mahajan> <[email protected]>

* offline tools rpc response

Signed-off-by: <Keval Mahajan> <[email protected]>

* offline tools rpc response message

Signed-off-by: <Keval Mahajan> <[email protected]>

* minor changes

Signed-off-by: <Keval Mahajan> <[email protected]>

* minor flake8 issues

Signed-off-by: <Keval Mahajan> <[email protected]>

* db migration for postgresql

Signed-off-by: <Keval Mahajan> <[email protected]>

* Check for existing tables in migration script
Signed-off-by: Madhav Kandukuri <[email protected]>

* db migration from status to enabled and reachable

Signed-off-by: <Keval Mahajan> <[email protected]>

* reachable value is updated as necessary

Signed-off-by: <Keval Mahajan> <[email protected]>

* Updated test cases and linting

Signed-off-by: Keval Mahajan <[email protected]>

* Use bootstrap script
Signed-off-by: Madhav Kandukuri <[email protected]>

* linting

Signed-off-by: Keval Mahajan <[email protected]>

* Use bootstrap script in main for table creation
Signed-off-by: Madhav Kandukuri <[email protected]>

* Revert "Use bootstrap script in main for table creation"

This reverts commit 20fcceb.

* Revert "linting"

This reverts commit 2ecc431.

* Revert "Use bootstrap script"

This reverts commit d579e8e.

---------

Signed-off-by: <Keval Mahajan> <[email protected]>
Signed-off-by: Keval Mahajan <[email protected]>
Co-authored-by: Madhav Kandukuri <[email protected]>

Author: kevalmahajan

alembic/versions/e4fc04d1a442_add_annotations_to_tables.py

@@ -14,6 +14,7 @@
 
 # Third-Party
 import sqlalchemy as sa
+from sqlalchemy.orm import Session
 
 # revision identifiers, used by Alembic.
 revision: str = 'e4fc04d1a442'
@@ -30,6 +31,14 @@ def upgrade() -> None:
     table. It includes a server-side default of an empty JSON object ('{}') to ensure
     that existing rows get a non-null default value.
     """
+    bind = op.get_bind()
+    sess = Session(bind=bind)
+    inspector = sa.inspect(bind)
+
+    if not inspector.has_table("gateways"):
+        print("Fresh database detected. Skipping migration.")
+        return
+
     op.add_column('tools', sa.Column('annotations', sa.JSON(), server_default=sa.text("'{}'"), nullable=False))
 
 
@@ -40,4 +49,12 @@ def downgrade() -> None:
     This function provides a way to undo the migration, safely removing the
     'annotations' column from the 'tool' table.
     """
+    bind = op.get_bind()
+    sess = Session(bind=bind)
+    inspector = sa.inspect(bind)
+    
+    if not inspector.has_table("gateways"):
+        print("Fresh database detected. Skipping migration.")
+        return
+    
     op.drop_column('tools', 'annotations')

alembic/versions/e75490e949b1_add_improved_status_to_tables.py

@@ -0,0 +1,46 @@
+"""Add enabled and reachable columns in tools and gateways tables and migrate data (is_active ➜ enabled,reachable).
+
+Revision ID: e75490e949b1
+Revises: e4fc04d1a442
+Create Date: 2025‑07‑02 17:12:40.678256
+"""
+
+# Standard
+from typing import Sequence, Union
+
+# First-Party
+# Alembic / SQLAlchemy
+from alembic import op
+
+# Third-Party
+import sqlalchemy as sa
+
+# Revision identifiers.
+revision: str = "e75490e949b1"
+down_revision: Union[str, Sequence[str], None] = "e4fc04d1a442"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade():
+    """
+    Renames 'is_active' to 'enabled' and adds a new 'reachable' column (default True) 
+    in both 'tools' and 'gateways' tables.
+    """
+    op.alter_column('tools', 'is_active', new_column_name='enabled')
+    op.add_column('tools', sa.Column('reachable', sa.Boolean(), nullable=False, server_default=sa.true()))
+
+    op.alter_column('gateways', 'is_active', new_column_name='enabled')
+    op.add_column('gateways', sa.Column('reachable', sa.Boolean(), nullable=False, server_default=sa.true()))
+
+
+def downgrade():
+    """
+    Reverts the changes by renaming 'enabled' back to 'is_active' 
+    and dropping the 'reachable' column in both 'tools' and 'gateways' tables.
+    """
+    op.alter_column('tools', 'enabled', new_column_name='is_active')
+    op.drop_column('tools', 'reachable')
+
+    op.alter_column('gateways', 'enabled', new_column_name='is_active')
+    op.drop_column('gateways', 'reachable')

mcpgateway/admin.py

@@ -712,7 +712,7 @@ async def admin_toggle_tool(
     form = await request.form()
     activate = form.get("activate", "true").lower() == "true"
     try:
-        await tool_service.toggle_tool_status(db, tool_id, activate)
+        await tool_service.toggle_tool_status(db, tool_id, activate, reachable=activate)
     except Exception as e:
         logger.error(f"Error toggling tool status: {e}")
 

mcpgateway/db.py

@@ -323,7 +323,8 @@ class Tool(Base):
     annotations: Mapped[Optional[Dict[str, Any]]] = mapped_column(JSON, default=lambda: {})
     created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utc_now)
     updated_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utc_now, onupdate=utc_now)
-    is_active: Mapped[bool] = mapped_column(default=True)
+    enabled: Mapped[bool] = mapped_column(default=True)
+    reachable: Mapped[bool] = mapped_column(default=True)
     jsonpath_filter: Mapped[str] = mapped_column(default="")
 
     # Request type and authentication fields
@@ -1025,7 +1026,8 @@ class Gateway(Base):
     capabilities: Mapped[Dict[str, Any]] = mapped_column(JSON)
     created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utc_now)
     updated_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utc_now, onupdate=utc_now)
-    is_active: Mapped[bool] = mapped_column(default=True)
+    enabled: Mapped[bool] = mapped_column(default=True)
+    reachable: Mapped[bool] = mapped_column(default=True)
     last_seen: Mapped[Optional[datetime]]
 
     # Relationship with local tools this gateway provides

mcpgateway/federation/forward.py

@@ -126,7 +126,7 @@ async def forward_tool_request(self, db: Session, tool_name: str, arguments: Dic
         """
         try:
             # Find tool
-            tool = db.execute(select(DbTool).where(DbTool.name == tool_name).where(DbTool.is_active)).scalar_one_or_none()
+            tool = db.execute(select(DbTool).where(DbTool.name == tool_name).where(DbTool.enabled)).scalar_one_or_none()
 
             if not tool:
                 raise ForwardingError(f"Tool not found: {tool_name}")
@@ -208,7 +208,7 @@ async def _forward_to_gateway(
         """
         # Get gateway
         gateway = db.get(DbGateway, gateway_id)
-        if not gateway or not gateway.is_active:
+        if not gateway or not gateway.enabled:
             raise ForwardingError(f"Gateway not found: {gateway_id}")
 
         # Check rate limits
@@ -263,7 +263,7 @@ async def _forward_to_all(self, db: Session, method: str, params: Optional[Dict[
             ForwardingError: If all forwards fail
         """
         # Get active gateways
-        gateways = db.execute(select(DbGateway).where(DbGateway.is_active)).scalars().all()
+        gateways = db.execute(select(DbGateway).where(DbGateway.enabled)).scalars().all()
 
         # Forward to each gateway
         results = []
@@ -292,7 +292,7 @@ async def _find_resource_gateway(self, db: Session, uri: str) -> Optional[DbGate
             Gateway record or None
         """
         # Get active gateways
-        gateways = db.execute(select(DbGateway).where(DbGateway.is_active)).scalars().all()
+        gateways = db.execute(select(DbGateway).where(DbGateway.enabled)).scalars().all()
 
         # Check each gateway
         for gateway in gateways:

mcpgateway/main.py

@@ -937,7 +937,7 @@ async def create_tool(tool: ToolCreate, db: Session = Depends(get_db), user: str
         logger.debug(f"User {user} is creating a new tool")
         return await tool_service.register_tool(db, tool)
     except ToolNameConflictError as e:
-        if not e.is_active and e.tool_id:
+        if not e.enabled and e.tool_id:
             raise HTTPException(
                 status_code=status.HTTP_409_CONFLICT,
                 detail=f"Tool name already exists but is inactive. Consider activating it with ID: {e.tool_id}",
@@ -1060,7 +1060,7 @@ async def toggle_tool_status(
     """
     try:
         logger.debug(f"User {user} is toggling tool with ID {tool_id} to {'active' if activate else 'inactive'}")
-        tool = await tool_service.toggle_tool_status(db, tool_id, activate)
+        tool = await tool_service.toggle_tool_status(db, tool_id, activate, reachable=activate)
         return {
             "status": "success",
             "message": f"Tool {tool_id} {'activated' if activate else 'deactivated'}",

mcpgateway/schemas.py

@@ -382,7 +382,8 @@ class ToolRead(BaseModelWithConfigDict):
     Includes all tool fields plus:
     - Database ID
     - Creation/update timestamps
-    - Active status
+    - enabled: If Tool is enabled or disabled.
+    - reachable: If Tool is reachable or not.
     - Gateway ID for federation
     - Execution count indicating the number of times the tool has been executed.
     - Metrics: Aggregated metrics for the tool invocations.
@@ -402,7 +403,8 @@ class ToolRead(BaseModelWithConfigDict):
     auth: Optional[AuthenticationValues]
     created_at: datetime
     updated_at: datetime
-    is_active: bool
+    enabled: bool
+    reachable: bool
     gateway_id: Optional[str]
     execution_count: int
     metrics: ToolMetrics
@@ -855,7 +857,8 @@ class GatewayRead(BaseModelWithConfigDict):
     - Database ID
     - Capabilities dictionary
     - Creation/update timestamps
-    - Active status
+    - enabled status
+    - reachable status
     - Last seen timestamp
     - Authentication type: basic, bearer, headers
     - Authentication value: username/password or token or custom headers
@@ -876,7 +879,9 @@ class GatewayRead(BaseModelWithConfigDict):
     capabilities: Dict[str, Any] = Field(default_factory=dict, description="Gateway capabilities")
     created_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc), description="Creation timestamp")
     updated_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc), description="Last update timestamp")
-    is_active: bool = Field(default=True, description="Is the gateway active?")
+    enabled: bool = Field(default=True, description="Is the gateway enabled?")
+    reachable: bool = Field(default=True, description="Is the gateway reachable/online?")
+
     last_seen: Optional[datetime] = Field(default_factory=lambda: datetime.now(timezone.utc), description="Last seen timestamp")
 
     # Authorizations