DADL: A Declarative Description Language for Enterprise Tool Libraries in LLM Agent Systems
New YAML format reduces tool catalog overhead from 142K to 1K tokens.
The Model Context Protocol (MCP) currently forces enterprises to deploy a dedicated server process for every API integration, each with its own dependencies and credential handling. Research shows these are often thin wrappers around REST APIs. Compounding the problem, MCP's per-tool registration model causes context window consumption to grow linearly with catalog size, forcing real-world deployments to expose only a fraction of available APIs.
DADL solves both issues by providing a single YAML file that describes endpoints, authentication, pagination, response shaping, and access classification. A shared runtime interprets these files at runtime, eliminating per-API servers and centralizing credentials. The Code Mode interface delivers the entire tool catalog to the LLM in a fixed-size context (about 1,000 tokens) regardless of catalog size, compared to 142,000 tokens with MCP. The v0.1 specification is open (CC BY-SA 4.0), and a public registry already contains 1,833 tool definitions across 20 services. This enables versioned, auditable enterprise tool libraries that any team can extend and share through a single auth boundary.
- Achieves 142x reduction in context cost: from ~142,000 tokens to ~1,000 tokens for tool advertisement.
- Public registry includes 1,833 tool definitions across 20 services, all describable in a single YAML file.
- Eliminates per-API server deployments and centralizes credential management through a shared runtime layer.
Why It Matters
Enables enterprises to scale LLM agent tool integrations without context window limits or deployment overhead.