Cryptographic agent identity. Attenuated authority. Entity resolution. One library across Rust, TypeScript, and Python - with byte-identical outputs.
cargo add kanoniv-agent-authnpm install @kanoniv/agent-authpip install kanoniv-agent-authimport { McpProof, verifyMcpCall } from "@kanoniv/agent-auth";
function handleToolCall(args: Record<string, unknown>) {
const { proof, cleanArgs } = McpProof.extract(args);
if (proof) {
const result = verifyMcpCall(proof, rootIdentity);
console.log(`Agent ${result.invoker_did} verified (depth: ${result.depth})`);
}
// use cleanArgs for your tool logic
}No external service. No API key management. The proof is self-contained - the server verifies the entire delegation chain using only the cryptographic material in the request.
Authority flows from a root (human or system) to agents, narrowing at each step:
Root (Human)
|-- delegates to Manager: [resolve, search, merge]
|-- delegates to Worker: [resolve] (narrower)
|-- calls MCP tool with proof
|-- server verifies entire chain back to rootCaveats accumulate. Each delegation can restrict actions, set expiry times, cap costs, or scope to specific resources. You can only give away a subset of what you have.
| Caveat | What it restricts |
|---|---|
action_scope | Allowed actions (e.g. ["resolve", "search"]) |
expires_at | RFC 3339 expiry timestamp |
max_cost | Cost ceiling for the operation |
resource | Resource glob pattern (e.g. "entity:customer:*") |
context | Key/value context match (e.g. session_id) |
custom | Arbitrary key/value constraint |
Adding auth to an MCP server? Start with MCP Server Auth - 5 lines, 5 minutes.
Building agents that delegate to each other? See Agent Identity then Delegation.
Need identity resolution for multi-agent systems? Go to the API Quickstart - resolve your first entity in 60 seconds.
Agent identity and delegation are MIT-licensed and open source at github.com/kanoniv/agent-auth. The entity resolution engine is a managed service at kanoniv.com.