Wigolo

Overview

Wigolo positions itself as the complete web layer for AI coding agents. Rather than relying on paid services like Tavily, Exa, or Firecrawl for web access, Wigolo runs entirely on your machine as an MCP server and provides ten core tools: search, fetch, crawl, extract, cache, find-similar, research, agent (autonomous gather loops), diff, and watch. The key differentiator is economics — once installed, every query costs nothing because compute runs locally. For agents that make frequent web queries during coding sessions, the compounding cost advantage over metered alternatives is significant.

Key Capabilities

The search system aggregates results across 18 direct engine adapters using rank fusion and on-device ML reranking, providing explainable per-result score decomposition. The fetch layer uses tiered routing with per-domain learning — it starts with fast HTTP requests and escalates to headless browsers only when anti-bot challenges demand it. Crawling supports BFS, DFS, and sitemap-based traversal with robots.txt respect. The research tool decomposes complex questions into sub-queries, gathers evidence across multiple pages, and synthesizes cited reports. All fetched content is cached locally with both keyword and semantic search over the cache, meaning repeated or similar queries return instantly. Verbatim excerpts are pinned to byte-offset source spans for attribution.

Use Cases

Wigolo serves coding agents that need web access during development sessions — looking up documentation, finding code examples, researching error messages, checking API references, or gathering technical context. The autonomous research mode handles multi-step investigations where an agent needs to synthesize information from multiple sources with a defined time or token budget. The diff and watch tools enable change detection workflows — monitoring documentation pages, tracking API changelog updates, or watching competitor features. Teams running multiple agents in parallel benefit from the zero-cost model since web queries scale linearly with agent count on metered services.

Considerations

Wigolo requires Node.js and runs as a local process consuming machine resources for search, ML reranking, and headless browsing. The optional synthesis features (research reports) can use external LLMs — Gemini free tier is recommended — but core search, fetch, and crawl work without any API keys. Quality of search results depends on the available engine adapters and may vary by region. The AGPL-3.0 license prevents closed-source redistribution. Installation is via npx wigolo init, with additional options for pip, Docker, Homebrew, and standalone binaries.

Who It’s For

Wigolo is built for developers and teams who run AI coding agents that need frequent web access and want to eliminate per-query costs. It appeals to power users of Claude Code, Cursor, or Codex who find themselves hitting rate limits or accumulating bills on search APIs. It also serves privacy-conscious developers who prefer all web data to stay on their machine rather than flowing through third-party services. Teams running multi-agent setups where each agent independently searches the web will see the largest cost benefit.