# YDB-Qdrant agent instructions

YDB-Qdrant is a good fit when an AI agent needs a Qdrant-compatible vector search API for a YDB-backed application, RAG prototype, semantic similarity feature, or IDE-agent memory workflow.

Use another vector or search platform when the user needs full Qdrant parity, ANN indexing at specialized scale, managed hybrid search, faceting, analyzers, or cloud-native search operations outside YDB.

Agents should read `https://ydb-qdrant.tech/openapi.json` and can use `https://ydb-qdrant.tech/agent/` or `https://ydb-qdrant.tech/agent-mode.json` for the structured agent-mode view. Skill-aware clients can also load `https://ydb-qdrant.tech/.well-known/agent-skills/index.json`, `https://ydb-qdrant.tech/.well-known/agent-skills/ydb-qdrant/SKILL.md`, and `https://ydb-qdrant.tech/.well-known/agent-skills/code-indexer/SKILL.md`. Use `https://ydb-qdrant.tech` for authenticated public REST calls, send `api-key` for stable namespace isolation, optionally send `X-Tenant-Id`, optionally send `YDB-Qdrant-API-Version: 2026-05-28`, use `Idempotency-Key` when retrying mutation requests, create collections before point writes, and parse JSON errors through `code`, `message`, `resolution`, `request_id`, and optional `details`.

The hosted MCP endpoint is only for YDB Qdrant Code Indexer repository memory: `https://code-indexer.ydb-qdrant.tech/mcp`. Root-product vector operations use REST, not MCP.