JACS

Portable cryptographic signatures for AI agents, services, and the artifacts they exchange.

JACS signs canonical JSON and common artifact formats, then lets Rust, Python, Node.js, Go, CLI, MCP clients, and other systems verify who signed what without a central server. Its schemas define verifiable JSON document formats so data can move between libraries, languages, and use cases without losing integrity.

cargo install jacs-cli | brew install jacs

What JACS does

Capability	What it means
Agent identity	Generate and manage a persistent cryptographic identity for an agent. Post-quantum ready (pq2025 / ML-DSA-87) by default.
Portable signatures	Sign in one surface and verify in another across Rust, Python, Node.js, Go, CLI, and MCP integrations.
Schema-backed JSON	Create verifiable JSON documents with declared schemas, content hashes, signer identity, signing algorithm, and signature metadata.
Artifact provenance	Sign files, Markdown/text, images, and Rust email payloads so consumers can detect tampering and identify the signer.
Local trust	Verify other agents with local keys, DNS anchors, and explicit trust policies (open, verified, strict).
Developer integration	Use the CLI, built-in MCP server, Rust crate, Python package, Node package, or Go bindings.

Quick start

cargo install jacs-cli

export JACS_PRIVATE_KEY_PASSWORD='your-password'
jacs quickstart --name my-agent --domain example.com
jacs document create -f mydata.json
jacs verify signed-document.json

Or via Homebrew:

brew tap HumanAssisted/homebrew-jacs
brew install jacs

This installs a single jacs binary with the CLI and MCP server built in.

Sign and verify more than JSON

JACS started with signed JSON documents and agent state. The same trust model now covers common AI-era artifacts:

Artifact	Interface	Notes
JSON and files	jacs document create, jacs verify, sign_message, sign_file	Self-contained signed envelopes for durable records, configs, memories, reports, and audit artifacts.
Markdown and text	jacs sign-text, jacs verify-text; Rust/Python/Node/Go bindings	Appends a readable JACS signature block to the file. Multi-signer review works without sidecar JSON.
Images	jacs sign-image, jacs verify-image; Rust/Python/Node/Go bindings	Embeds provenance in PNG, JPEG, or WebP metadata. Consumers verify signer identity and pixel-content integrity.
Email	Rust jacs::email	Signs raw RFC 5322 .eml bytes by adding a jacs-signature.json MIME attachment, then verifies field-level content hashes.

These signatures prove that a given agent signed specific canonical bytes at its claimed time. They do not prove first creation, copyright ownership, or real-world authorship by themselves.

MCP server

JACS includes a stdio-only MCP server for Claude Desktop, Cursor, Claude Code, Codex, and other MCP clients:

jacs mcp

{
  "mcpServers": {
    "jacs": {
      "command": "jacs",
      "args": ["mcp"]
    }
  }
}

The MCP server opens no HTTP port. It runs as a subprocess of the MCP client so the agent private key stays local to that process.

Core profile (default) includes state, document, trust, audit, memory, search, and key tools.

Full profile (jacs mcp --profile full) adds agreements, messaging, A2A, and attestation tools.

Use cases

Local provenance — Create, sign, verify, and export agent documents locally. No server required.

Reviewable text — Let multiple agents or reviewers counter-sign a README, design doc, policy, or release note in place.

Media provenance — Attach verifiable signer identity to photos, charts, screenshots, or AI-generated images without a sidecar file.

Email provenance — Add a JACS signature attachment to raw email and verify important headers, body parts, and attachments.

Agent boundaries — Sign tool outputs, API responses, MCP calls, A2A artifacts, or multi-agent agreements when data crosses a trust boundary.

Platform verification — For verified documents, agent behavior, benchmarks, and hosted workflows around JACS identities, see HumanAssisted/haiai.

When you do not need JACS

• Everything stays inside one service you control and logs are enough.
• You only need accidental-corruption detection; a checksum is simpler.
• There is no meaningful trust boundary or audit requirement.

JACS is most useful when signed data leaves the process, service, team, or organization that produced it.

Language support

The CLI and MCP server are the recommended starting points. Native APIs are available when you need direct library integration:

Language	Install	Notes
Rust	cargo add jacs	Deepest API surface, including jacs::email, jacs::text, and jacs::media.
Python	pip install jacs	Simple API, framework adapters, text/image signing.
Node.js	npm install @hai.ai/jacs	Async-first API, framework adapters, text/image signing.
Go	go get github.com/HumanAssisted/JACS/jacsgo	Signing and verification bindings for services.

Security

• Private keys are encrypted with password-based key derivation.
• The MCP server is stdio-only and opens no network listener.
• Signatures include algorithm identification and downgrade protection.
• Automated tests cover cryptographic operations, password validation, agent lifecycle, DNS verification, media/text signing, and attack scenarios.
• pq2025 (ML-DSA-87 / FIPS-204) is the default signing algorithm for new agents.

Report vulnerabilities to security@hai.ai. Do not open public issues for security concerns.

Links

• Documentation
• Quick Start Guide
• Inline Text Signatures
• Image and Media Signatures
• Email Signing and Verification
• Development Guide
• HAI.AI Platform
• HAI SDK

---

v0.11.1 | Apache-2.0 | Third-Party Notices