Query

Syntheia Query is an API and Model Context Protocol (MCP) interface that provides structure-aware retrieval over legal documents for teams building software and AI workflows on top of legal text. Rather than retrieving passages by keyword or vector similarity, it navigates a document's clause hierarchy, defined terms, and cross-references, returning only the clauses relevant to a query along with the definitions and cross-referenced provisions those clauses depend on. Each result is linked back to its source document and location, supporting review and verification of downstream output.

Key features and functions include:

Structure-aware retrieval
Query navigates the actual structure of a legal document — its clause hierarchy, defined terms, and cross-references — rather than matching text by keyword or embedding similarity. It enables an AI agent to return the clauses that bear on the question, not the passages that happen to resemble it.

Resolved references
Results include whatever the returned clauses depend on.  Where a clause turns on a defined term, or is qualified by a cross-reference, those are returned with it, so the retrieved material is complete enough to reason over without further round trips.

Source-linked results
Every returned clause is tied back to its source document and its location within that document.  Output can be traced to the underlying text, supporting review, citation and verification of anything built on top of it.

API and MCP access
Query is available as an API for direct integration into applications and pipelines, and over the Model Context Protocol (MCP) for agents and AI assistants to call as a tool.  Teams can build against it without implementing document parsing, structural analysis or retrieval logic themselves.

Token efficiency
Because retrieval is targeted rather than exhaustive, Query passes substantially fewer tokens to a model than supplying full documents, while retaining the same level of accuracy.  This makes document-heavy AI workflows materially cheaper to run at volume.