Tools & Agents

The Tools & Agents view shows analytics on how tools and agents are being used across your organization. Use it to understand your MCP server footprint, identify overused or anomalous tools, and track agent deployment coverage.

Navigate to Highflame StudioObservatoryTools & Agents.


Tool usage analytics

The tool usage panel shows aggregate statistics on tool invocations across all agent sessions:

Metric
Description

Total calls

Number of tool calls in the selected time range

Unique tools

Number of distinct tools called

Success rate

Fraction of tool calls that completed without error

Block rate

Fraction of tool calls blocked by a guardrail or gateway policy

P50 / P95 latency

Tool call latency percentiles

Top tools

A ranked list of tools by call volume, with per-tool breakdown of success rate, block rate, and average latency. Click any tool to filter the event table below to calls for that tool.

Tool category distribution

A breakdown of tool calls by category (file system, web search, code execution, database, external API, MCP). This helps identify whether your agents are concentrating usage in high-risk categories.


MCP server visibility

If you are using the Agent Gateway with MCP routing, the MCP panel shows:

Column
Description

MCP server

The MCP server name and endpoint

Tools exposed

Number of tools registered on this server

Calls (7d)

Total calls routed to this server in the past 7 days

Agents using

Number of distinct agents that have called tools on this server

Policy

The credential and access policy applied to this server

Status

Whether the server is active and reachable

Click any MCP server to see its full tool list and call history.


Agent inventory

The agent panel lists all agent identities that have been active in the selected time range:

Column
Description

Agent ID

The ZeroID agent identity

Display name

Human-readable name from the ZeroID registration

Sessions (7d)

Number of sessions in the past 7 days

Tool calls (7d)

Total tool invocations

Unique tools

Number of distinct tools the agent has called

Violations (7d)

Threat events associated with this agent

Evasion rate

Fraction of events that were blocked vs. total (higher = more aggressive agent)

Products

Which Highflame products have observed this agent

Risk score

UEBA-derived composite risk score

Last seen

Most recent activity timestamp

The evasion rate and product breadth (number of distinct products the agent has been seen across) are the two strongest UEBA signals for identifying agents behaving anomalously. An agent with a high evasion rate is generating a disproportionate number of blocks relative to its peers. An agent appearing across many products in a short window may indicate credential reuse or identity spoofing.

Agents with elevated risk scores or high violation counts are highlighted. Click any agent to filter the Threats and Sessions views to that agent.


Anomaly detection

Observatory continuously monitors tool and agent behavior for anomalies relative to each entity's baseline. The baseline is computed from a rolling 7-day window using z-score analysis — the same method used by the detector drift heatmap in Command Center.

Anomaly types:

  • New tool access — an agent calls a tool it has not used before

  • Unusual call volume — a tool is called significantly more than its baseline rate (z-score flagged)

  • New destination — a tool makes an outbound call to a domain not seen before

  • Off-hours activity — an agent is active outside its normal operating hours

  • Cross-agent reuse — the same credential or session token is observed across multiple agent identities

  • IDE shift — an agent that normally runs in one IDE is suddenly active in a different one

  • Sudden product breadth increase — an agent accesses a new Highflame product it has not been seen in before

Anomalies surface as annotations in the tool and agent tables and as events in the Threats view with the source observatory:anomaly.

Last updated