The shortlist
Models train on a snapshot; your dependencies don't stop moving. Two MCP servers fix the two halves of this — current library docs, and real web search — and a third lets you self-host the search side entirely.
These are documentation and catalogue-based suggestions, not hands-on benchmarks. Check the linked product sources, supported languages, regional availability, permissions and current costs. Use the same evaluation worksheet for each candidate.
63k stars · Open source · TypeScript
Library-documentation lookup through MCP and a CLI with skills. Check the package version and returned official page; retrieving documentation does not guarantee a correct generated implementation.
Capability source ↗— Your stack is niche enough that its docs are not covered.
5.1k stars · Open source · TypeScript
Search built for AI consumption, exposed over MCP — one config entry gives your agent a real search tool instead of stale training data.
1.3k stars · Open source · TypeScript
An MCP bridge to a SearXNG instance. Self-hosting the bridge does not keep search queries away from upstream search engines; inspect the complete query path and logging configuration.
— You have no SearXNG instance to point at; the server requires an instance URL.
Every retrieval tool you add widens what the agent reads — and web content is untrusted input. Prefer read-only search/docs servers, and treat 'the agent found it online' with the same skepticism you'd give any unverified source.
