What is Agnost AI?

Analytics and observability for AI agents and MCP servers

Agnost AI gives you visibility into what your AI agents are doing — every tool call, session, latency, error, and user interaction — tracked automatically and surfaced in a real-time dashboard.

How it works

Your AI Agent / MCP Server
        ↓
  Agnost AI SDK  (one line of code)
        ↓
  Agnost AI Backend  (api.agnost.ai)
        ↓
  Dashboard  (app.agnost.ai)

Every tool call your agent makes is captured as an event inside a session. Sessions group all events from a single conversation. You can filter by user, tool, time range, success rate, and latency — and run LLM-as-a-judge classification on top.

Key concepts

Concept	Description
Org ID	Your workspace identifier. All events are scoped to an org. Get it from app.agnost.ai.
Session	A single conversation or agent run. Groups related events together.
Event	A single tool/function call within a session. Has a name, latency, success/failure, and optional input/output.
Checkpoint	A named timing point inside an event — useful for profiling multi-step tools.
User	An optional identity attached to a session via user_id for per-user analytics.

MCP Servers

One-line integration. Wrap your MCP server and every tool, resource, and prompt call is tracked automatically.

TypeScript

MCP server in TypeScript

Python FastMCP

MCP server with FastMCP

Go

MCP server in Go

Anthropic (low-level)

Low-level Anthropic Python SDK

Conversational Agents

For non-MCP agents. Use begin() / end() to wrap any AI interaction and track it manually.

Python

Python conversation SDK

TypeScript

TypeScript conversation SDK

Integrations

Apify Actors

Apify cloud actors

MCP Toolbox

Google MCP Toolbox for Databases