Local-first MCP server for agent web research and extraction
wigolo, from KnockOutEZ, is a local-first Model Context Protocol server that supplies AI coding agents with on-device web research and retrieval capabilities. The tool performs deep site crawling, structured extraction, and multi-engine searching to feed agents with searchable context. Key functionality includes crawling, on-device indexing, and agent-facing MCP tools for integration. It targets developers and AI researchers who need private, searchable web memory to augment code-focused agents and reduce reliance on external search APIs.
What tasks can you actually use it for?
The tool maps to research and data-retrieval steps in an agent workflow: it can crawl documentation sites, extract tabular and metadata, and produce indexable records for agents to query. Practical tasks include gathering API docs for code generation, extracting structured tables from reference pages, and building a local searchable memory that agents consult during prompt resolution. It exposes standardized MCP tools so agents can request fetched excerpts and metadata directly.
How consistent and explainable are the search results?
Results are ranked with on-device machine learning models, producing explainable relevance scores rather than opaque rankings. The server stores a persistent local cache and a SQLite-based vector memory to serve repeated queries instantly. For complex pages the tiered fetcher escalates into a headless browser to capture dynamically rendered content, which improves coverage but increases processing time and local resource use.
What inputs and environment does it require?
Deployment requires a Node.js 20+ environment and can be started via npx or run inside Docker. The server accepts remote crawling targets, sitemap mapping, and explicit seed URLs, and supports structured extraction modes driven by JSON Schema as well as CSS selector and table-specialized extraction. Compatibility with MCP-enabled clients is explicit, so non-MCP integrations require an adapter or a REST API bridge.
How does it fit into developer workflows and toolchains?
The tool plugs into agent-driven development by offering rank fusion across multiple search adapters and persistent local memory for offline re-queries. Integration is easiest for projects already using MCP-enabled clients; the developer ecosystem includes explicit compatibility with agents such as Claude Desktop and Cursor. Teams that need cloud-free indexing can operate entirely on local hardware and keep fetched content on-device for later agent use.
Practical choice for privacy-minded developers who adopt MCP
The tool is a practical choice for developers and researchers who prioritize local control and agent integration, backed by its open-source origins and visible community interest. It demands hands-on setup and MCP-aware clients, so it best serves teams willing to operate a local server and adapt agent workflows. For those constraints, it provides a viable path to keep agent web access private and under direct control.
