Best for
- MCP server developers
- Teams building Claude Connectors or ChatGPT Plugins
- Teams needing tool usage analytics
- Organizations seeking simple telemetry integration
A
38FMRS · D
AgentCat provides a powerful and easy-to-integrate analytics platform for MCP server owners, capturing user intent and behavior with a simple one-line integration. It supports popular tools like OpenTelemetry, Datadog, Sentry, and PostHog, and offers advanced identity and redaction features. While free for open source, commercial users may need to consider costs. Overall, it's a valuable observability solution.
AgentCat is an analytics platform for MCP server owners. It captures user intentions and behavior patterns to help you understand what AI users actually need from your tools, eliminating guesswork and accelerating product development with a single line of code. This SDK also provides a free and simple way to forward telemetry like logs, traces, and errors to any OpenTelemetry collector or popular tools like Datadog, Sentry, and PostHog. AgentCat works as lightweight middleware inside your MCP server, injecting session_id, agent_id, and context parameters, along with an additional get_more_tools tool, to enable agent session replay, trace debugging, and support for existing observability platforms.
Tools
track
Track your MCP server with AgentCat, injecting analytics parameters and enabling telemetry exporters
identify
Provide identity for each caller, linking events to a consistent user or entity
redactSensitiveInformation
Apply custom redaction to text data before sending to AgentCat to remove sensitive information
redactEvent
Modify or drop events at the event level by returning a modified event or null
Setup
- Sign up at agentcat.com and obtain your project ID.
- Install the npm package in your MCP server project: npm install agentcat
- Import agentcat and call agentcat.track(server, projectId) in your server code.
- Optionally configure identify, redactSensitiveInformation, or redactEvent hooks, and exporters to integrate with existing observability platforms.
- Call track() in the correct lifecycle location (e.g., inside factory functions) for your MCP server architecture.
Fit and risk
Not for
- Developers not using MCP
- Simple MCP servers with no need for analytics or telemetry
- Scenarios with strict privacy requirements against third-party analytics services
Required permissions
- Requires ability to modify MCP server tool schemas and inject additional parameters
- Sends telemetry data and session replays to AgentCat servers or configured exporters
- May access HTTP authentication information and headers for user identification
Risks and side effects
- Adds context overhead via injected parameters, but less than 1%
- Sending telemetry to third-party services may raise data privacy concerns, though AgentCat redacts sensitive data and encrypts at rest
- Depends on AgentCat service availability, but diagnostics failures do not affect tool calls
- Free for open source, but commercial use may incur costs
Troubleshooting
- Ensure the project ID is correct and valid
- Check network connectivity so SDK can reach agentcat.com or configured telemetry endpoints
- Review local ~/agentcat.log for diagnostic information
- Ensure track() is called on the correct server instance, especially for stateless servers
- To disable diagnostics, set disableDiagnostics: true or environment variable DISABLE_DIAGNOSTICS
Use cases
See how agents use your MCP server and uncover missing features
Debug sessions where agents get stuck or confused
Forward analytics to OpenTelemetry, Datadog, Sentry, or PostHog
Collect user intent data to guide product development
Supported clients
Claude DesktopFull support