ADL vs MCP (Model Context Protocol)
Short answer: ADL and MCP address different layers and are designed to work together. MCP (Model Context Protocol) standardizes how a model or agent connects to tools, data sources, and context through MCP servers. ADL (Agent Definition Language) describes what an agent is — its identity, permission boundaries, security, lifecycle, and compliance — and governs how those tools may be used. An ADL document can generate MCP server configurations. MCP gives the agent reach; ADL gives it identity and guardrails.
What is MCP?
MCP (Model Context Protocol), introduced by Anthropic, is an open protocol for connecting AI models and agents to tools, resources, and prompts through a standard client–server interface. An MCP server exposes capabilities (tools to call, resources to read, prompts to use) that any MCP-compatible model can consume, so integrations are written once and reused across models.
What is ADL?
ADL is an open standard for defining an AI agent in a single machine-readable document: identity, capabilities and tools, permission boundaries, security requirements, data classification, lifecycle, and compliance posture — the agent's auditable "passport." Where MCP describes how to reach a tool, ADL declares which tools an agent has, what it is permitted to do with them, and who authorized it.
Key differences
| MCP | ADL | |
|---|---|---|
| Primary purpose | Connect models to tools, data, and context | Define an agent's identity and governance |
| Layer | Tool / context protocol | Identity, permissions, compliance |
| Identity | Not an identity model | Cryptographic identity (DID) + attestation |
| Permissions | Transport for tool access | Deny-by-default boundaries over that access |
| Governance / compliance | Out of scope | NIST 800-53, SOC 2, ISO 27001, EU AI Act mappings |
| Lifecycle | Out of scope | Status, versioning, sunset dates, successors |
| Relationship | Provides tool access | Generates MCP server configurations |
How ADL and MCP work together
ADL sits around MCP. You declare an agent's tools, resources, and prompts — plus the permissions and security that bound them — in one ADL document, then generate MCP server configurations from it. Per the ADL specification §15.2, an implementation maps ADL fields to an MCP configuration: name, description, version, tools, resources, and prompts.
The result: the tools an agent reaches over MCP are the same ones declared in its governed ADL passport, inside the permission boundaries ADL enforces — so capability and control stay aligned.
When to use which
- Use MCP when you need an agent to call tools and read data through a standard interface.
- Use ADL when you need a portable, auditable definition of the agent's identity, permissions, and compliance — and to generate MCP configurations from it.
- Use both together for capable agents whose tool access is also governed and audit-ready.
FAQ
Is ADL a replacement for MCP?
No. MCP is a protocol for connecting to tools and context; ADL is a definition and governance layer. ADL complements MCP and can generate MCP server configurations.
Can I use ADL and MCP together?
Yes — that's the intended pattern. Declare the agent's tools and permissions in ADL, then generate its MCP configuration so tool access and governance share one source of truth.
Does ADL define the tool protocol?
No. MCP remains the protocol for tool and context access; ADL declares which tools an agent has and the permissions around them.