Obsigna — sign and verify every AI agent action
Obsigna is the tooling layer for the Agent Receipts protocol — cryptographically signed, tamper-evident audit trails for every action your AI agent takes.
The protocol spec lives at agentreceipts.ai. This site covers the concrete tools: how to install them, configure them, and start generating receipts.
Choose your path
Section titled “Choose your path”Hook — the fastest path
Section titled “Hook — the fastest path”The obsigna-hook binary integrates directly with agent runtimes (Claude Code PostToolUse, others). Records native tool calls before they reach any MCP layer.
Best for: Claude Code users, runtime-level audit requirements.
→ Hook overview
→ Claude Code integration
MCP Proxy — no code changes
Section titled “MCP Proxy — no code changes”The proxy sits in front of any MCP server and automatically signs a receipt for every tool call, without modifying the client or server.
Best for: Claude Desktop, Claude Code, Cursor, Windsurf, VS Code Copilot, Cline, Codex — any MCP-connected agent.
→ MCP Proxy overview
→ Quick Start
SDKs — emit receipts from your code
Section titled “SDKs — emit receipts from your code”Use the Go, TypeScript, or Python SDK to emit receipts directly from your application. Full control over what gets recorded, when, and how.
Best for: custom agents, server-side code, automated pipelines.
→ Go SDK · TypeScript SDK · Python SDK
What you get
Section titled “What you get”Every tool call produces a receipt: a W3C Verifiable Credential signed with Ed25519, hash-chained into a tamper-evident sequence. The signing keys and receipt store live in the obsigna-daemon — outside the agent process.
The chain is independently verifiable with obsigna receipt verify. See the Agent Receipts protocol for the full specification.