Open-Agent-DB is an offline-first catalog of agent skills, MCP servers, and directives, searchable locally by keyword (SQLite FTS5) and semantic vectors.
Yes, Open-Agent-DB is open source under the MIT license.
Open-Agent-DB is free to use.
Yes, Open-Agent-DB can be self-hosted (the source is available under the MIT license).
Open-Agent-DB is an offline-first catalog and search engine for AI agent skills, MCP servers, and project directives, queryable via npx open-agent-db search with full-text (SQLite FTS5) and dense-vector semantic search. Open source: yes (MIT); self-hostable; pricing free. It runs on local SQLite, is offline-first, and launched 2026-09-28.
Open-Agent-DB is a searchable index of the things agents are built from — skills, MCP servers, and project directives — that lives on your machine. You query it from the command line with npx open-agent-db search, and it answers two ways: keyword search over a local SQLite FTS5 index, and dense-vector semantic search for meaning-based matches. Because it is offline-first, the catalog is a local snapshot you can query without a network round-trip, and it ships with a paired Hugging Face dataset (ns0bj/open-agent-db) behind it. Open source: yes (MIT); self-hostable; pricing free.
As a GTM stack leans on agents, "what tool or skill already exists for this" becomes a real discovery problem — and hosted directories cost a network hop and a trust dependency. Open-Agent-DB puts a searchable catalog of skills, MCP servers, and directives on your own machine, with both keyword and semantic search from the CLI, so an engineer (or an agent) can find capabilities offline. Open source: yes (MIT); self-hostable; pricing free. The honest read: the offline-first, dual-search design is a genuinely useful discovery primitive, but the ~3.48M-asset / 113k-server index counts are the maker's own figures, the local snapshot can drift out of date, and this is a 0-star launch-day project — treat it as a fast local index to refresh, not an authoritative live registry.