Search¶

PriorisMCP exposes retrieval mechanisms over content already fetched into its storage, covered as separate chapters in this section:

• Full-text search — SearchIndex, literal/lexical keyword matching over persisted chunks/leaves (v1).
• Vector search — VectorSearchBackend, embedding-based semantic search over the same corpus (v3).
• Graph search — GraphSearchBackend, a single-layer knowledge graph over Pointer/Concept nodes, plus a NetworkX algorithm layer on top.

Full-text and vector search share one tool, research_search_fetched's mode parameter (see Vector search → Composition), because both dispatch to the same contract: a query string in, a uniform ranked-matches list out. Graph search does not share that contract — get_node returns one node, neighbors returns edge/node pairs, subgraph returns a whole node/edge set — so it is exposed as its own three tools (research_graph_write/research_graph_query/research_graph_analyze) instead, not a fourth mode value — see Graph search → research_graph_* are new tools, not a mode on research_search_fetched.

Not to be confused with Discovery: all three mechanisms above operate over content already fetched into this project's storage. Discovery is a remote, third-party mechanism for finding candidates not yet fetched at all — see Discovery for the exact boundary.