Framework Integrations
First-class adapters for the agent frameworks developers actually use.
Framework Integrations
Checkrd ships first-class adapters for every major agent framework. Each adapter targets the framework's canonical extension point — callback handlers, tracing processors, hooks, or transport subclasses — so policy enforcement plugs in idiomatically without monkey-patching.
Python
| Framework | Extension point | Install |
|---|---|---|
| LangChain / LangGraph | BaseCallbackHandler | pip install 'checkrd[langchain]' |
| OpenAI Agents SDK | TracingProcessor + InputGuardrail | pip install 'checkrd[openai-agents]' |
| Anthropic Claude Agent SDK | PreToolUse / PostToolUse hooks | pip install 'checkrd[claude-agent-sdk]' |
| Model Context Protocol (MCP) | ClientSession subclass + handler wrap | pip install 'checkrd[mcp]' |
The Python SDK also ships transparent vendor instrumentors for OpenAI, Anthropic, Cohere, Mistral, Groq, Together, and Google GenAI. They sit at the HTTP transport layer and work alongside the framework adapters above without double-counting.
JavaScript / TypeScript
| Framework | Extension point | Install |
|---|---|---|
| LangChain.js | BaseCallbackHandler | npm install checkrd @langchain/core |
| OpenAI Agents (TypeScript) | TracingProcessor + InputGuardrail / OutputGuardrail | npm install checkrd @openai/agents |
| Claude Agent SDK (TypeScript) | PreToolUse / PostToolUse hooks via attachToOptions | npm install checkrd @anthropic-ai/claude-agent-sdk |
| Vercel AI SDK | LanguageModelV2Middleware | npm install checkrd ai |
| Mastra | wrapMastraAgent proxy + Telemetry provider | npm install checkrd @mastra/core |
| Model Context Protocol (MCP) | wrapMcpClient / wrapMcpServer proxies | npm install checkrd @modelcontextprotocol/sdk |
| Next.js | App Router and Pages Router middleware | npm install checkrd |
| Hono | app.use(...) middleware | npm install checkrd hono |
| Cloudflare Workers | Edge-runtime withCheckrd HOC | npm install checkrd |
The JS SDK also ships vendor instrumentors for OpenAI, Anthropic, Cohere, Mistral, Groq, Together, and Google GenAI.
Why these specific extension points
Patching SDK internals (the "monkey-patch every vendor's HTTP client" approach) breaks on every minor release of every framework. Each adapter Checkrd ships uses the framework's explicitly documented public extension point — the contract the framework author committed to.
| Framework | Equivalent products that use this surface |
|---|---|
| LangChain callbacks | Langfuse, Logfire, OpenLLMetry |
| OpenAI Agents tracing | Langfuse, Logfire |
| Claude Agent hooks | Anthropic-recommended pattern |
| MCP handler wrap | Standard practice (no native middleware) |
If a framework you use is not listed, open an issue — most adapters land in under 200 lines because the contract is small.