Compass developer docs

The explainable dietary decision layer for AI agents.

Quickstart

Make a first Search request with curl, Node, or Python.

MCP server

Install Compass in Claude Desktop, Cursor, or Codex.

API reference

Read endpoint contracts, examples, fields, and error behavior.

Transparency

Review source tiers, evidence policy, and conservative output rules.

Which surface should I use?

Field	Type	Notes
MCP server	Agent tools	Start here when your host can install MCP, such as Claude Desktop, Cursor, or Codex.
REST / OpenAPI	Production apps	Use this when your backend needs exact HTTP contracts, auth behavior, errors, and response schemas.
SDK examples	Convenience clients	Use SDK-style examples for faster setup, while treating OpenAPI as the source of truth.

Field	Type	Notes
Search	Find candidates	Use when you have a dietary query and need ranked restaurant results.
Enrich	Add Compass context	Use when you already have a restaurant or place record from your product.
Decision	Return a verdict	Use when your product needs fit, not_fit, or unknown for a user profile.
Restaurant Details	Supporting lookup	Use after Search or Enrich when you need the full structured record for one result.

Start here

Compass exposes three v1 surfaces: restaurant search, restaurant enrichment, and dietary fit decisions. Responses include reason codes, evidence pointers, confidence, and verification guidance.

Read the quickstart