The protocol layer for AI Agents accessing data — it tells the Agent what it can read and write before execution, rather than letting the Agent guess.
The Agentic Data Protocol (ADP) is an open protocol that sits between AI agents and your data systems. Before an agent ever touches data, ADP exposes a precise, policy-governed contract — describing exactly which resources exist, what fields they contain, and what operations are permitted. Agents discover, describe, and validate their intent before execution, eliminating guesswork and unsafe free-form queries.
- Safe by design — Policy manifests govern which agents can access which resources, with validation before execution
- Deterministic — Structured Intent IR replaces free-form queries, making agent data access auditable and reproducible
- Heterogeneous — One protocol, one SDK, many backends; no per-database glue code in your agent
- 📐 Read the Specification for protocol details and schema
- 🖥️ Deploy the ADP Hypervisor to serve your data backends
- 🐍 Use the Python SDK to connect agents to any ADP-compliant server
- 🔌 Use the ADP Connectors to bridge ADP with LLM orchestration frameworks
| Repository | Description |
|---|---|
| agenticdataprotocol | Protocol specification, JSON Schema, and TypeScript types |
| adp-hypervisor | Reference server — translates ADP intents to backend queries, enforces policies |
| python-sdk | Python client SDK for connecting agents to any ADP-compliant server |
| adp-connectors | Connectors for LLM frameworks — MCP bridge, Copilot skills, and more |
We welcome contributions of all kinds — bug fixes, new backends, SDK improvements, and documentation. Please open an issue or pull request in the relevant repository to get started.
Have questions? Join the conversation in our community discussions.
ADP is open source, licensed under the Apache 2.0 License.