Company profile · 3 min read
AgentCat shows teams exactly how agents use MCP servers and connected products
AgentCat provides agent session replay, goals, issue detection, analytics and experiments for MCP servers, Claude Connectors and ChatGPT Apps.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
AgentCat is for teams whose software is becoming an agent interface and whose existing analytics cannot show what those agents actually did. It provides session replay, agent goals, issue detection and performance data for MCP servers, Claude Connectors and ChatGPT Apps.
What it does
AgentCat’s current site says teams can step through every tool call in a session, inspect requests, responses and errors, group sessions by the goal the agent was pursuing, and track per-tool performance and failure patterns. The product also advertises A/B testing, Slack triage and an open-source SDK. AgentCat homepage
The practical advantage is that it treats an agent as a user with a path, not only a server with logs. A product team can see that an agent repeatedly failed at one tool, misunderstood a goal or reached a dead end, then decide whether to change the tool contract, the product flow or the source content. The docs say enriched telemetry can be forwarded to Datadog, Sentry and PostHog through OpenTelemetry. AgentCat docs
Price and tradeoffs
The Free plan is $0 and includes 500 sessions per month, up to three teammates, replay and the analytics dashboard. Growth is listed at $160 per month for 2,000 sessions, unlimited projects and up to ten teammates; Enterprise is custom with exports, SSO/SAML, audit logs and SLA guarantees. Agent Goals is an add-on at $90 per month for 1,000 classifications, with overage listed separately. AgentCat pricing
The company says it is SOC II compliant and offers redaction hooks, self-hosting and enterprise controls. Those are company statements; no security or privacy test was performed. The main product tradeoff is telemetry depth versus data exposure. Prompts, tool calls and file-related events can contain sensitive context, so the team needs a clear redaction, retention and access policy before instrumenting production.
The Speedrun profile identifies Kashish Hora, formerly in AI B2B at Grammarly and Superhuman, and Naseem Alnaji, a prior Opal Security founder.
My editorial take
I would start AgentCat when agent sessions are already producing support tickets or roadmap arguments that logs cannot resolve. The Free tier is enough to inspect the shape of the problem. Move to Growth when volume and team collaboration justify it; do not buy the goal-classification add-on until raw replays have shown that intent grouping will change a product decision.
Quick facts
| Field | Sourced detail |
|---|---|
| Buyer | Teams building MCP servers, connectors and agent-facing apps |
| Product | Agent session replay, goals, issues, analytics and experiments |
| Pricing | Free to 500 sessions; Growth $160/month; Enterprise custom |
| Data control | Redaction hooks, self-hosting and enterprise controls advertised |
| Main fit question | Can you connect agent behavior to a concrete product decision? |
Sources checked
AgentCat’s Speedrun profile, homepage, pricing page and docs were checked on 2026-09-19.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗