Three steps to your
first signed receipt.
You do not need to rewrite your agents or prompts. Wrap an existing MCP server or add our lightweight middleware to your agent framework.
Wrap any MCP server (Claude Code, Cursor, Claude Desktop)
Prefix your MCP server command with hvr mcp wrap (bound to publishing-targets.json). Every tool call can be evaluated against your policy and saved as a signed .hvr receipt.
Verify the receipt in your browser or terminal
Drop the generated .hvr file onto the Verify page or run the open-source hvr CLI verifier with zero network calls.
Add to your agent framework or API gateway
Adapters are available for TypeScript, Python, Go, and Rust, plus drop-in middleware for popular agent frameworks. Publishing targets are bound in publishing-targets.json and activate after repositories and packages are public.
Supported frameworks & gateways.
Vercel AI SDK & LangChain.js
Wrap tool definitions with withHivemindReceipts() to attach a verified receipt ID to every tool execution.
LangGraph, CrewAI, AutoGen, LlamaIndex & OpenAI Agents
Use the @receipt_tool decorator or callback handler built from source.
Envoy, Kong, Cloudflare Worker & OpenTelemetry
Issue receipts at the network edge or convert OpenTelemetry gen_ai.* spans into signed receipts automatically.
Compliance evidence mapping.
Record-keeping & human oversight
Maps signed tool call receipts and passkey approval records to automatic event logging and human-in-the-loop oversight controls.
ICT incident & change audit trail
Maps sequence-complete agent logs and emergency halt receipts to operational resilience evidence.
Access & AI system change records
Maps policy evaluation digests and checkpoint inclusion proofs to logical access and change-management controls.