Open websites to AI agents under owner-controlled policy.

Documentation: ajarprotocol.org

Most agents reach websites by pretending to be hurried users: load the page, scrape the HTML, submit a form, maybe carry the user's browser session. That is bad for everyone. The agent cannot tell public data from account data, the owner cannot publish a clear policy, the user cannot delegate only a narrow job, and a purchase or account change leaves weak evidence. Ajar Protocol gives websites a cleaner path: signed manifests, semantic Views, typed Actions, scoped Mandates, SIMULATE before execution, and dual-signed Receipts.

Ajar is a protocol layer over HTTPS for agent-readable content, typed actions, delegated authority, and verifiable receipts. A website owner publishes a signed manifest that tells agents what is available, who may use it, at what price, and under what proof of authority.

Ajar does not replace HTML, checkout systems, payment rails, MCP, or agent frameworks. It gives websites a standard way to declare and govern agent access while integrating with the protocols companies already use.

Today, agents often treat websites as screens to scrape. That is expensive, fragile, and unsafe for real actions. Ajar adds the missing owner-side contract:

1. A site owner installs or runs an Ajar Gateway, plugin, or native integration.
2. The owner reviews generated Views, Actions, policy, pricing, and gates.
3. The owner signs a Capability Manifest at /.well-known/ajar.json.
4. An agent Kernel discovers the manifest, verifies signatures, reads Views, simulates actions, checks mandates, commits only when allowed, and stores receipts.

The owner key is the root of authority. Automation drafts; owners decide.

Published now:

Publishing later:

Ajar is designed for adoption by companies, platforms, hosts, agencies, and independent implementers. Reference implementations get no special status: compatibility comes from the public conformance suite.

Ajar is intended to work alongside:

• MCP and WebMCP for tool/action transport
• ACP and UCP for commerce semantics
• x402, AP2, and MPP for settlement
• A2A for agent-to-agent workflows
• RFC 9421 / Web Bot Auth for signed agent identity
• UCAN-style chains for capability delegation

Project contact: Hardik Modha hardik@technopers.com

• Specification and documentation: CC-BY-4.0
• Code and machine-readable implementation artifacts: Apache-2.0
• Long-term governance target: neutral standards venue

Contributors should read CONTRIBUTING.md. Automation contributors must follow AGENTS.md.

Phase 0 is complete: the spec v0.1 draft baseline is frozen and the independent-reader exit gate passed (record). Phase 1, the Gateway MVP read layer, is underway — the Gateway skeleton (T1.1), the executable conformance harness (T1.14), and the docs MCP server have shipped, and an agent-built demo site already passes the conformance CORE probes through the reference Gateway.

See the ROADMAP and BUILD-ORDER in planning.

• New here? Follow ONBOARDING.md end to end.
• Read the ajar README in its stated order.
• Pick a task from the org Project board, Ajar Roadmap.
• Start with issues labeled good first issue when learning the system.
• Follow the org CONTRIBUTING.md: one task, one PR, DoD demonstrated.
• AI-agent contributors must follow AGENTS.md.