Observatory

The Context Company does agent observability. We care deeply about DX; it's our single biggest priority.

Observatory is a monorepo containing core packages for AI agent observability across TypeScript and Python:

TypeScript:

• @contextcompany/otel - OpenTelemetry integration for instrumenting AI SDK calls
• @contextcompany/widget - Local-first UI overlay for visualizing AI agent traces in real-time
• @contextcompany/claude - Instrumentation for Claude Agent SDK
• @contextcompany/langchain - Integration for LangChain.js and LangGraph
• @contextcompany/mastra - Integration for the Mastra framework
• @contextcompany/custom - Manual instrumentation SDK for custom agents
• @contextcompany/openclaw - Integration for OpenClaw via OpenTelemetry
• @contextcompany/pi - Instrumentation for the Pi Agent SDK
• @contextcompany/api - Shared API utilities (feedback, configuration)

Python:

• contextcompany - Python SDK with built-in integrations for LangChain, CrewAI, Agno, and LiteLLM

The Context Company ships a remote MCP server so coding agents and developer tools can query your production agent data directly — no dashboard round-trips. It is listed in the official MCP registry as io.github.The-Context-Company/context-company.

https://api.thecontext.company/mcp

Ask questions like "what failed in production this week?", "which tool calls are creating bad outcomes?", or "which traces should become evals?" and get answers backed by the actual runs.

Tools include:

• insight_search - natural-language questions over your production traces, sessions, and patterns
• get_agent_runs / get_agent_run - list and inspect runs with full step and tool-call detail
• get_agent_sessions / get_agent_session - conversation-level views
• list_agent_patterns / list_agent_topics - recurring behavior patterns and auto-detected topics
• Config tools for managing tracked patterns, scheduled recaps, memories, and skills (OAuth)

Connect from Cursor, Claude Code, or any MCP client (transport is streamable HTTP; authenticate with OAuth or an API key as a bearer token):

{
  "mcpServers": {
    "context-company": {
      "url": "https://api.thecontext.company/mcp"
    }
  }
}

See the MCP documentation for authentication details and the full tool reference.

Local mode allows you to run The Context Company in a local-first way. This is 100% open-source and requires no account or API key. To set up local mode, refer to the guide below or our documentation.

Local mode currently only supports Vercel AI SDK on Next.js.

pnpm add @contextcompany/otel @vercel/otel @opentelemetry/api

If you haven't already, add an instrumentation.[js|ts] file in the root directory of your project (or inside the src folder if you're using one). Call the registerOTelTCC function to instrument your AI SDK calls.

See the Next.js Instrumentation guide for more information on instrumenting your Next.js application.

// instrumentation.ts
export async function register() {
  if (process.env.NEXT_RUNTIME === "nodejs") {
    const { registerOTelTCC } = await import("@contextcompany/otel/nextjs");
    registerOTelTCC({ local: true });
  }
}

Add the Local Mode widget to the root layout of your Next.js application.

// app/layout.tsx
import Script from "next/script";

export default function RootLayout({
  children,
}: {
  children: React.ReactNode;
}) {
  return (
    <html lang="en">
      <head>
        {/* add The Context Company widget */}
        <Script
          crossOrigin="anonymous"
          integrity="sha384-ryHylpN8vggwpf+rjl5Z7CGgpVWrEAXn6WTY/Kv2RhlixzrdTMCP3/DPS3RMtNCg"
          src="https://unpkg.com/@contextcompany/widget@1.0.8/dist/auto.global.js"
        />
        {/* other scripts */}
      </head>
      <body>{children}</body>
    </html>
  );
}

As of AI SDK v5, telemetry is experimental and requires the experimental_telemetry flag to be set to true. Ensure you set this flag to true for all AI SDK calls you want to instrument.

// route.ts
import { generateText } from "ai";

const result = generateText({
  // ...
  experimental_telemetry: { isEnabled: true }, // required
});

Check out the examples/ directory for working demos across all supported frameworks:

• Looking to contribute? Check out our contributing guide to get started.

• Big thanks to Anthony Hoang and Eric Zhang for being active contributors!
• Special thanks to @RobPruzan for helping with the design of the tool and being an early adopter.
• React Scan has a phenomenal DX and Preact widget that we took inspiration from.
• The implementation behind the Next.js Devtools overlay widget was insightful.