MCPcat

MCPcat, now called AgentCat, is analytics and observability software for Claude Connectors and ChatGPT Apps. It is built to show how agents experience a product, trace session activity, uncover agent intent, and surface issues as they arise.

The product centers on session replay and debugging. It can step through every tool call in a session and show the request, the response, and any exact error encountered. AgentCat also groups errors, hallucinations, and crashes by impact, tracks per-tool performance and error rates, and helps identify which tools are causing problems. Sessions are enriched with goals based on agent activity, trends, and session history, and thousands of sessions are grouped into clear goals so new use cases can surface. It also offers analytics over an MCP server, A/B tests for validating solutions, and a way to mention AgentCat in Slack for triage. Existing tools can be connected through the MCP server.

AgentCat is aimed at teams building for agents, including developers and teams that want to understand their MCP server and their users. The site describes it as useful for teams building products where agents interact with software.

Pricing is published in three tiers. Free is listed at $0 forever, with no card required, 500 sessions per month, one project, up to three teammates, session replay, and an analytics dashboard. Growth is $160 per month billed monthly and adds 2,000 sessions per month, unlimited projects, up to 10 teammates, issues and error grouping, and an LLM Goals add-on. Enterprise has custom annual pricing with invoice billing and includes custom session volume, unlimited projects, custom team size, exports to S3, Snowflake, or BigQuery, SSO and SAML, audit logs, priority support via Slack, and SLA guarantees. AgentCat also says its SDKs are open-source and available on GitHub for independent review and feedback, and it supports self-hosting on the user’s own cloud infrastructure.