mulesoft · valkyrie69 · Dec 4, 2025 · Dec 4, 2025 · Dec 4, 2025 · Dec 5, 2025
@@ -26,9 +26,9 @@ For more information, see xref:view-and-navigate-agent-network.adoc#viewing-node
 Edges represent the relationships and communications among your nodes:
 
 * Declared edges appear between node instances when no runtime information is available because there's no registered traffic between them. Dashed lines on the graph indicate declared edges.
-* Runtime edges appear between nodes when traffic is registered and a Flex Gateway exists. Solid lines on the graph indicate runtime edges. Clicking on a runtime edge opens the details card and shows the relevant metrics.
+* Runtime edges appear between nodes when traffic is registered and a managed Flex Gateway exists. Solid lines on the graph indicate runtime edges. Clicking on a runtime edge opens the details card and shows the relevant metrics.
 +
-Edges are drawn under specific conditions, primarily when the calling agent is protected by a managed Flex Gateway at egress. For more information, see xref:gateway::flex-gateway-managed-ingress-egress.adoc[].
+Edges are drawn under specific conditions, primarily when the calling agent is protected by a managed Flex Gateway at egress. For agent-to-MCP relationships, edges are automatically drawn when the calling agent’s egress or the MCP server’s ingress is protected by a managed Flex Gateway. You don't need to add new policies or extra configuration to make edges visible. For more information, see xref:gateway::flex-gateway-managed-ingress-egress.adoc[]
 
 * If an agent has an egress managed Flex Gateway, when it calls another agent, the edge is visible.
 * If the requesting agent doesn't have an egress Flex Gateway, no edge metric is generated and the edge isn't shown.
@@ -54,7 +54,6 @@ Agent Visualizer provides a single, consistent architectural view of all assets
 
 For more information about switching between production and sandbox environment layers, see xref:view-and-navigate-agent-network.adoc#view-data-flow-layers[View Data Flow Layers].
 
-
 [[details-card]]
 == Details Cards
 
@@ -66,7 +65,7 @@ For more information, see xref:view-and-navigate-agent-network.adoc#view-node-de
 
 === Node Metadata
 
-When you select a node, the details card shows metadata for that asset, including: 
+When you select a node, the details card shows metadata for that agentic asset, including: 
 
 [%header,cols="30a,70a"]
 |===
@@ -123,6 +122,24 @@ Each instance entry includes the asset version number, so you can easily disting
 +
 Instances are ordered from the most recent (latest) to least recent (earliest) asset version, making it easy to track version history over time.
 
+=== MCP Server Details
+
+The MCP server details card includes active MCP tools and per‑instance activity:
+
+* Active tools reflect the selected Activity Period and only include tools invoked during that window.
+* Tools are aggregated across all instances of the MCP server, ranked by throughput, with the top 10 displayed by default. Select *Show More* to view additional tools.
+
+[NOTE]
+Tool visibility requires the MCP asset to be published in Exchange and telemetry from a managed Flex Gateway on the relevant path. Runtime usage shows only allowlisted tools that have been called.
+
+=== Tool Details
+
+When you select a tool, the tool details card opens. It shows:
+
+* Tool name, icon, and MCP asset information at the top.
+* Active instances count (instances with traffic for this tool). If more than two instances are shown, a scrollbar is enabled.
+* A list of instances with per‑instance metrics (total calls and error rate) for the selected tool.
+
 [[policy-visualization]]
 == Policy Visualization
 

@@ -25,7 +25,7 @@ For more information, see xref:anypoint-code-builder::af-define-your-agent-netwo
 
 [[declared-edge]]
 Declared edge::
-A design‑time relationship between two nodes (from the agent network YAML or Exchange metadata) indicating an intended connection when no runtime traffic is observed; it’s rendered as a dashed line.
+A design‑time relationship between two nodes (from the agent network YAML or Exchange metadata) indicating an intended connection. When no runtime traffic is observed, it’s rendered as a dashed line.
 +
 For more information, see xref:agent-visualizer-components.adoc#edges[Edges].
 

@@ -1,12 +1,12 @@
 = Agent Visualizer Overview
 
-Agent Visualizer provides a detailed visual graph so you can see, understand, and interact with your enterprise's evolving network of agents, Model Context Protocol (MCP) servers, and large language models (LLMs). Agent Visualizer gives you insight into hidden agent interactions, governance gaps, and performance bottlenecks. 
+Agent Visualizer provides a detailed visual graph of your agent network so you can see, understand, and interact with your enterprise's evolving network of agents, Model Context Protocol (MCP) servers, and large language models (LLMs), including agents created in external platforms like Bedrock, Copilot, and Vertex AI. Agent Visualizer gives you insight into hidden agent interactions, governance gaps, and performance bottlenecks. 
 
 
 Agent Visualizer is an intuitive user interface that helps you:
 
 * Gain complete situational awareness of your agentic ecosystem with a view of all connected agents in your agent network on one map, including clear relationships and direction of calls.
-* See unclear interaction patterns by showing real-time call volume for the past 24 hours. This helps you understand key pathways and identify under- or overused connections. 
+* See unclear interaction patterns by showing real-time call volume. This helps you understand key pathways and identify under- or overused connections. 
 * See governance gaps with a view of which agents are protected by Flex Gateways and policies. 
 * Understand node purpose and origin.
 + 

@@ -12,7 +12,7 @@ include::partial$partial-agent-visualizer.adoc[tag=agent-visualizer-permissions]
 +
 On the Agent Visualizer page, you see the agent network map showing all discovered agentic assets, nodes and their relationships (edges).
 
-If the page is empty, there are no discoverable agentic assets in Anypoint Exchange. For more information about creating an agent network project and publishing assets to Exchange, see xref:anypoint-code-builder::af-create-agent-network.adoc[].
+If the page is empty, there are no discoverable agentic assets in Anypoint Exchange (Exchange). For more information about creating an agent network project and publishing assets to Exchange, see xref:anypoint-code-builder::af-create-agent-network.adoc[].
 
 [[navigate-agent-network]]
 == Navigate the Agent Network
@@ -55,6 +55,143 @@ When you select an edge, the details card displays relationship information and
 . Select an edge to open its details card.
 . Examine the traffic throughput per hour. 
 
+[[view-external-nodes]]
+== View External Agents and MCP Servers
+
+External agents and MCP servers that are published to Exchange appear as nodes in the agent network. You can select these nodes to view metadata and, when available, instance details.
+
+. Select an external agent or MCP server node to open its details card and review the node’s metadata.
++
+To visualize runtime edges for external agents, see xref:view-standalone-agents[View Agents Without Flex Gateway].
+
+[[view-mcp-server-edges-tools]]
+== View MCP Server Edges and Tools
+
+When agents invoke MCP servers, Agent Visualizer shows the relationship on the graph and lists MCP tools on the MCP server node’s details card.
+
+. Select an MCP server node to focus associated edges and open its details card.
++
+Tool invocations from multiple agents are aggregated into a single list on the MCP server:
++
+** Tools are ordered by throughput (total MCP calls).
+** The top 10 tools are shown by default. 
++
+Click *Show More* to see additional tools when available.
++
+[NOTE]
+For tool visibility, the MCP server must be present in Exchange and the egress path must be protected by a managed Flex Gateway. This applies to MuleSoft‑hosted MCP servers defined in the broker network and externally hosted MCP servers defined in the broker network.
++
+The tools list reflects the selected activity period and shows only tools that were invoked during that window. Allowlisted tools that aren't called don't appear. If the MCP server exposes a tool catalog that Agent Visualizer can query, the details panel shows the full set of available (design‑time) tools alongside runtime usage.
++
+. To see traffic metrics on the connection, select the edge between the agent and the MCP server and open the edge details card.
+
+=== Example: Visualize Interactions Between External Agents and MCP Servers
+
+To visualize runtime edges for external agents and MCP servers that aren't part of a defined agent network, you must publish them to Anypoint Exchange and route their traffic through Flex Gateway. 
+
+This is an example of how to visualize the interactions between agents and MCP servers when they aren’t part of an agent network. This process manually "onboards" your external assets so they appear in the agent network graph as managed nodes with live traffic edges.
+
+. Publish your MCP servers to Anypoint Exchange.
++
+For more information about creating assets in Exchange, see xref:exchange::to-create-an-asset.adoc#create-mcp[].
+. Set up a Flex Gateway in Connected Mode.
++
+This gateway acts as the "exit point" that intercepts calls going to the external servers.
++
+For more information about securing agent interaction in Flex Gateway, see xref:gateway::flex-agent-secure.adoc[].
+. In API Manager, create a new instance for each external MCP server.
+.. Point the Implementation URI to the actual external URL of your MCP server.
+.. Deploy these instances to your Egress Flex Gateway.
+. Enable telemetry by injecting identifying headers for visualization (for example, source and destination identifiers):
+.. In API Manager, go to the *Policies* tab for these instances.
+.. Apply the Agent Connection Telemetry policy. This allows the gateway to parse MCP-specific traffic (like list tools).
+.. Record the Consumer URL (the Flex Gateway endpoint) for each MCP server. 
+. Publish the agent asset in Anypoint Exchange with the *Agent* type.
+. Setup the Ingress Flex Gateway to handle the inbound traffic to the agent.
+. In API Manager, go to *Agent and Tools Instances* and select *Add > Agent*. 
+.. Assign the agent to the Ingress Flex Gateway.
+.. Copy the Agent Instance ID. 
++
+The external agent uses this ID to identify itself to the Ingress Flex Gateway.
+. In API Manager, apply a *Header Injection* policy to the agent instance.
+.. Add the *Header Name* as `x-agent-instance-id` and the *Header Value* as the Agent Instance ID you copied earlier.
+.. Click *Apply*.
+. On the calling agent app, inject the `x-anypoint-api-instance-id` header (on list tools or call tools operations) providing the agent Instance ID from the ingress entry in API Manager as the value of the header.
+
+[[view-standalone-agents]]
+== View Agents Without Flex Gateway
+
+Agents that aren't protected by a managed Flex Gateway appear as standalone nodes. You can still view their metadata and understand their purpose in the network.
+
+. Select the agent node to open its details card.
++
+Review metadata, tags, and any instance details available.
+. To visualize runtime edges and performance metrics, protect the agent with a managed Flex Gateway on the relevant path (ingress or egress), then generate traffic.
+
+[[search-agent-network]]
+== Search the Agent Network
+
+Use search to locate nodes in the agent network. Search is case‑insensitive and supports multiple fields.  Search results reflect the current base graph. Environment layer selection (Production or Sandbox) affects instance‑level metrics but not which nodes can be found by search.
+
+. Open the search box.
++
+.Show me how
+[%collapsible]
+====
+* Click the magnifying glass icon.
+* Use the keyboard shortcuts:
+** Mac: Cmd+Shift+k
+** Windows: Ctrl+Shift+k
+====
+. Enter your search terms.
++
+You can search by:
++
+* Display name 
+* Type, for example, broker, agent, MCP 
+* Platform, for example, MuleSoft, Agentforce, Bedrock 
+* Exchange tags
++
+Smart suggestions show up to 10 results, ordered by relevance, with type and platform badges and an indicator of the matched field. 
+. Select a result to zoom in on the node and open its details card.
+. Press the escape key to clear the results and close the search box.
+
+== Add Filters
+
+Use filters to control which parts of the network you see on the canvas. Filters help you focus on the information you need in large graphs by limiting the scope to specific business groups and time windows. 
+
+The agent network graph shows assets that belong to the selected business groups, and edges between those assets (including cross‑business group connections) are shown on the same map.
+
+Metrics on nodes and edges are aggregated based on the business groups you select. When you deselect business groups, metrics update to reflect only the remaining selection.
+
+[[filter-business-groups]]
+=== Apply Filters to the Agent Network
+
+. Open the filter panel and select the filters to apply:
++
+* *Business Groups*
++
+By default, your current business group is selected. You can select multiple business groups to see cross-business group connections in a single view. Only business groups you have Agent Visualizer permissions for appear in the list.
++
+When multiple business groups are selected, the canvas shows assets belonging to the selected business groups and edges between those assets across all selected business groups. The metrics that are shown are aggregated across all selected business groups. When you deselect business groups, metrics update to reflect only the remaining selection.
++
+[TIP]
+If performance degrades when viewing multiple business groups at once, reduce the selection, use search, or zoom to areas of interest.
+* *Environment*
++
+Select *Production* or *Sandbox*.
+* *Activity Period*
++
+Select the last:
+** *24 hours*
+** *3 days*
+** *5 days*
+** *7 days*
+* *Display Options*
++
+You can hide standalone nodes and edges.
+. Click *Apply Filters*.
+
 == See Also
 
 * xref:index.adoc[]