What if your AI coding agent could search the web, fetch pages, crawl documentation sites, and synthesize research reports without burning through API credits or sending your data to a third party? Wigolo, an open-source Model Context Protocol server that hit public beta this month, answers that question with a local-first architecture that costs exactly zero dollars per query. At 986 GitHub stars and climbing, it is quietly becoming the default web intelligence layer for developers who want their agents to think without a meter running.
What Is Wigolo
Wigolo is a single Node.js process that exposes ten tools over MCP (Model Context Protocol), REST, and SDKs. It runs wherever your agent runs: as an MCP server next to Claude Code or Cursor in your terminal, as an HTTP endpoint inside Docker on your server, or embedded inside your own application via TypeScript or Python SDKs. Install it with one command, and your agent gains search, fetch, crawl, extract, cache, find-similar, research, agent (autonomous gather loops), diff, and watch capabilities. The core tools need no API keys, no cloud account, and no metered billing. Everything lives under ~/.wigolo/, including the cache, embeddings, models, and configuration.
The project was built by the solo developer @KnockOutEZ and is licensed under AGPL-3.0. It is deliberately designed to stay free: the expensive parts ranking, embeddings, and the browser engine run on your own hardware, so there is no per-query cost to recover. The README puts it plainly: sustained by donations, with the AGPL license legally preventing a bait-and-switch into a closed hosted product.
Ten Tools That Replace Cloud APIs
Wigolo's tool set covers virtually every web intelligence need an AI coding agent might have. The search tool runs 18 direct adapters across public search engines with rank fusion and ML reranking, producing an explainable per-result evidence score. Every search result carries a verbatim excerpt pinned to its exact byte position in the source, a citation ID the agent can quote, and a score the agent can inspect. Weak results are flagged as junk by the built-in scorer. Failed engines are reported. Stale cache is labeled. The agent always knows what it is standing on.
The fetch tool loads a URL through a tiered router that auto-escalates from plain HTTP to a headless browser engine when it encounters anti-bot challenges or JavaScript-heavy single-page applications. It returns clean markdown, metadata, and links. The crawl tool handles multi-page BFS, DFS, sitemap-based, or map-only crawls with per-domain rate limits and robots.txt respect. The extract tool pulls structured data from pages, tables, JSON-LD, metadata, brand identity, and named schemas like Article, Recipe, or Product. The cache tool lets agents query everything already seen by keyword or hybrid semantic search, with change detection across visits. And the research tool decomposes a question into sub-queries, fans out across sources, and synthesizes a cited report or a structured brief for the host LLM to assemble.
The autonomous agent tool runs a full gather loop: plan, search, fetch, extract, synthesize, with a step log, a time budget, and an optional output schema. It is essentially a research assistant that lives inside your coding agent and can run unsupervised tasks.
How It Compares to Paid Services
Wigolo's README includes a live benchmark where a single Claude Fable 5 session fanned out the same query to four web tools: built-in WebSearch, Wigolo, Tavily, and Exa. All four converged on the same core answer and the same top source. But only Wigolo returned verbatim excerpts pinned to byte-offset source spans, an explainable score decomposition, and live per-engine telemetry. When two of its results were weak, its own scorer flagged them as junk on-screen. The cloud tools earned their line too: Exa rendered the official documentation comparison matrix in full. The key differentiator is not quality; it is ownership. Wigolo keeps every query on your machine, caches everything, and costs nothing to re-query.
Firecrawl, Exa, and Tavily all require API keys and charge per query. Wigolo requires none. Firecrawl offers multi-engine search and whole-site crawling but lacks the local cache, the byte-pinned evidence, and the explainable scoring that Wigolo provides. Exa excels at deep extraction but is metered and cloud-only. Tavily is straightforward and API-friendly but again metered. For solo founders and small teams, the difference compounds fast: an agent that asks fifty questions an hour costs nothing with Wigolo and adds up quickly with any of the cloud alternatives.
Getting Started
Getting started takes exactly one command. Run npx wigolo init --agents=claude-code (or cursor, codex, gemini-cli, vscode, windsurf, zed) and the tool downloads the browser engine and on-device models, runs a health check, writes the MCP configuration, and prints a per-component summary. The install is unattended by default, safe for scripts and CI. Node 20 or higher and about 1.5 GB of free disk space are the only requirements. Once installed, the core tools search, fetch, crawl, extract, cache, and find-similar work with no API key at all. For the research and agent tools that use an LLM for synthesis, a free Gemini API key is recommended as the single biggest quality upgrade. The tool also supports Anthropic, OpenAI, Groq, or fully offline local models via Ollama.
From there, your agent can search the web, crawl documentation sites, extract structured data from competitor pages, cache everything it has seen for instant re-query, and run autonomous research loops. Every result is cached locally, so asking the same question twice costs nothing.
Who This Is For
Wigolo is built for solo founders, indie developers, and small engineering teams who run AI coding agents as part of their daily workflow. If you use Claude Code, Cursor, Codex, Gemini CLI, Windsurf, or any MCP-compatible agent and find yourself hitting API rate limits or burning through search budgets, Wigolo replaces that entire cost layer. It is also ideal for privacy-conscious teams who do not want their agent's queries sent to a third-party search API. The local cache makes it particularly useful for developers who frequently look up the same documentation, API references, or code patterns. The plugin system lets you add custom search engines or site extractors in roughly 100 lines, making it extensible for domain-specific workflows. For teams running self-hosted agents in n8n, LangChain, CrewAI, or LlamaIndex, the Docker image and SDKs drop Wigolo in as a local service that never phones home.