Get Started in 60 Seconds

Two ways to get started: Install via IDE (recommended) or Standalone (for any agent framework).

Install via IDE

For Cursor and Claude Code users. Get static analysis, dynamic runtime insights, and correlation directly in your IDE.

Claude Code

Run these commands:

1 /plugin marketplace add cylestio/agent-inspector
2 /plugin install agent-inspector@cylestio

Restart Claude Code for the MCP connection to activate.

Cursor

Run this command:

Fetch and follow instructions from https://raw.githubusercontent.com/cylestio/agent-inspector/main/integrations/AGENT_INSPECTOR_SETUP.md

Restart Cursor and approve the MCP server when prompted.

What you get with IDE integration
  • Persistent memory — Agent Inspector remembers context across sessions
  • Runtime insights — Access dynamic data from your running agents
  • Slash commands — /agent-scan, /agent-fix, /agent-gate
  • Correlation — Connect static code findings with runtime behavior

Standalone Installation

For any agent framework without IDE integration. Start the proxy and dashboard with one command.

For Anthropic (Claude)

uvx agent-inspector anthropic

For OpenAI (GPT)

uvx agent-inspector openai
What happens
  • Proxy starts on http://localhost:4000
  • Dashboard opens at http://localhost:7100
  • No installation needed — uvx handles everything
What is uvx?

uvx automatically downloads Agent Inspector, runs it in an isolated environment, and cleans up afterward. No pip install, no conflicts. Learn more

CLI Options

Customize Agent Inspector's behavior with these command-line options.

Option Description Default
--base-url Override the LLM provider base URL (e.g., for local proxy) Provider default
--ui-port Override the dashboard port 7100
--use-local-storage Enable persistent SQLite storage for trace data Disabled (in-memory)
--local-storage-path Custom database path (requires --use-local-storage) ./agent-inspector-trace.db
--log-level Set logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL) INFO
--no-presidio Disable Presidio PII detection Enabled

Example with Options

uvx agent-inspector anthropic --ui-port 8080 --use-local-storage

Point Your Agent to the Proxy

Change one line in your agent code:

Anthropic

client = Anthropic(
    api_key="your-key",
    base_url="http://localhost:4000"
)

OpenAI

client = OpenAI(
    api_key="your-key",
    base_url="http://localhost:4000"
)

Group Sessions with Agent Workflow ID

Optionally, include a workflow ID in the base URL to group sessions and link them to static analysis findings: 

http://localhost:4000/agent-workflow/{AGENT_WORKFLOW_ID}

Use the same agent_workflow_id when running static scans via MCP tools to correlate findings.

n8n users: Update your Anthropic/OpenAI credential's Base URL field to http://localhost:4000. See n8n guide →

Run Multiple Test Sessions

Execute your agent with varied inputs during evaluation and testing. Run a minimum of 5+ sessions, ideally more than 20, the more the better. Agent Inspector analyzes automatically as sessions complete.

for prompt in test_prompts:
    agent.run(prompt)  # Each execution = 1 session
Data Storage

Default: Data is in-memory only and cleared when you stop Agent Inspector.
Persistent: Use --use-local-storage to save data to SQLite and persist across restarts.

View Your Agent Assessment

Open the dashboard at http://localhost:7100 and immediately get:

  • Static Analysis — Code vulnerabilities mapped to OWASP LLM Top 10
  • Dynamic Analysis — Runtime behavior patterns and anomalies
  • Correlation — Static findings validated against runtime evidence
  • Behavioral Analysis — Stability scores, clustering, outlier detection
  • Production Gates — Go/no-go decision based on security posture
Agent Inspector Dashboard

IDE Slash Commands

If you set up IDE integration, use these commands directly in Cursor or Claude Code:

Command Description
/agent-scan Run static analysis on your agent code
/agent-analyze Run dynamic analysis on runtime behavior
/agent-fix Get AI-powered fix for a security finding
/agent-correlate Connect static and dynamic findings
/agent-gate Check production readiness status
/agent-report Generate compliance report
/agent-debug Explore sessions and debug agent behavior

What's Next?

Platform Guides

Security Lifecycle

Capabilities

Alternative Installation Methods

If you prefer traditional package managers:

# pipx (recommended for CLI tools)
pipx install agent-inspector
agent-inspector anthropic

# pip (standard Python)
pip install agent-inspector
agent-inspector anthropic
← Overview How It Works →