M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
W

Wigolo

Community
Local-first web intelligence for AI coding agents.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #158 of 438
Stars
★ 5.5k Very popular
Transport
stdio (local process) · SSE
Runtime
Node.js 20+ · Docker
Credentials
Optional API key
License
Other / unspecified
Last commit
Tools
10
57FMRS · C

The supplied material describes wigolo as a KnockOutEZ-maintained public-beta, local-first web-intelligence MCP server. It supports stdio MCP and exposes SSE for remote MCP clients through its server mode; core search, fetch, crawl, extract, cache, and similarity features require no API key. Research and autonomous-agent synthesis can optionally use an LLM provider.

Strongest · Documentation 15/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
11/20
Maintenance
10/20
Documentation
15/20
Setup experience
13/20
Why each score
Reliability 8/20
The README describes Node≥20, MCP over stdio, ten tools, doctor, and degradation behavior, but no server manifest, source, or verifiable test/CI files were supplied. The claimed 7,600-test suite and CI badge are not static evidence. A normal path is plausible, but MCP initialization, tool-list consistency, dependency failures, and key-path coverage cannot be verified, so substantial points are deducted.
Security and permissions 11/20
The evidence indicates local-by-default operation, storage under ~/.wigolo/, opt-in LLM and aggregator backends, bearer-token protection for non-loopback HTTP binding, and stated robots.txt/rate-limit behavior. Points are deducted because the service actively accesses the public web and supports authenticated sessions, page actions, proxies, and remote REST/MCP; no manifest, source, or permission model verifies credential isolation, confirmation of dangerous actions, request scoping, log redaction, or network boundaries. No stated red-line risk is established, but the evidence does not support a high score.
Maintenance 10/20
The README labels the project public beta and provides GitHub issue and contribution paths, an AGPL-3.0-only license, and a SECURITY.md channel, indicating an ownership and response path. Points are deducted because no commit history, release record, dependency-update evidence, actual issue response, or security-handling evidence was supplied; the input metadata also says NOASSERTION, conflicting with the README license claim.
Documentation 15/20
The README offers layered quickstart guidance, client coverage, tool summaries, configuration variables, REST/SDK/Docker/framework examples, troubleshooting, privacy/security documentation links, and disclosed limitations. Points are deducted because the referenced documentation, tool parameters, limits, implementation evidence, and contracts were not supplied; several capability, benchmark, and compatibility claims cannot be independently reviewed, and the README appears internally inconsistent about the tool count.
Setup experience 13/20
The README provides npx initialization commands, a Node≥20 prerequisite, mainstream client options, cross-platform support, doctor, and uninstall commands, making the path broadly clear. Points are deducted because installation downloads a browser and local models, requires about 1.5GB of disk, and may involve native builds, Linux libraries, and first-run downloads; no actual client configuration file, CI execution evidence, or successful connection log was supplied, so the static calibration prevents a higher score.

Static review · not runListed 2026-07-29

Read the FMRS scoring method →

Fit and risk

What it can accessWrites / deletes local filesUses the networkControls a browser

Best for

  • Developers who prioritize local processing and privacy
  • Agents that need web tooling without per-query charges
  • Teams needing search, crawling, caching, and research in one surface
  • macOS, Linux, or Windows users with about 1.5 GB of available disk space

Not for

  • Users seeking a fully hosted service without local browser or model downloads
  • Workflows that require guaranteed access to every anti-bot or challenge-protected site
  • Environments that prohibit public-web crawling or optional third-party LLM synthesis
  • Users who need to keep a modified network-service version closed source

Required permissions

  • Local filesystem access is needed for cache, configuration, embeddings, and models under ~/.wigolo/.
  • Network access is needed to search and retrieve the public web.
  • Browser mode may start a local browser engine; Linux may require additional operating-system libraries.
  • Research, agent, and answer-format search can optionally call Gemini, Anthropic, OpenAI, Groq, Ollama, or another compatible provider.
  • Authenticated pages and page actions may require session information.

Risks and side effects

  • Public search engines and target sites may fail, become stale, rate-limit requests, or block access; wigolo reports degradation and challenge failures in its output.
  • Datacenter IP reputation can reduce access to challenge-protected sites.
  • Optional LLM synthesis sends relevant requests to the selected provider; Ollama or another compatible local endpoint can keep synthesis local.
  • Browser and model downloads require about 1.5 GB of disk space.
  • The project uses AGPL-3.0-only; modified versions run as network services must publish their modifications.

Setup

Before you start

Runtime：Node.js 20+ · Docker

GEMINI_API_KEY optionalsecret Free Gemini API key from aistudio.google.com/apikey; upgrades research/agent/search format=answer into a finished cited answer.
WIGOLO_GITHUB_TOKEN optionalsecret Optional GitHub token raising GitHub code search from 10 to 30 req/min.
WIGOLO_API_TOKEN optionalsecret Bearer token required for HTTP server access from off-loopback (docker serve).
Other optional settings (6)
WIGOLO_LLM_PROVIDER optional Optional LLM provider for synthesis (gemini/anthropic/openai/groq/ollama); ollama stays keyless and local.
WIGOLO_SEARCH optional Set to hybrid to widen the retrieval funnel with an aggregator fallback engine.
WIGOLO_TLS_TIER optional Set to auto for per-domain learned fetch hardening.
WIGOLO_EAGER_WARMUP optional Set to 1 to pay the ~1s model load up front instead of lazily.
USE_PROXY optional Set to true to route through a proxy; pair with PROXY_URL.
PROXY_URL optional Proxy endpoint URL used when USE_PROXY=true; add NODE_EXTRA_CA_CERTS for TLS-inspecting proxies.

Install Node.js 20 or newer, then run npx wigolo init. The command sets up the local engine, downloads the browser engine and on-device models, and performs a health check. Use --agents=claude-code,cursor or another supported comma-separated list to wire agent configurations during setup. Other MCP clients can register npx -y wigolo in their own MCP configuration. Run npx wigolo doctor to check health.

Check that it works

Run `npx wigolo init` (optionally with --agents=claude-code,cursor), then `npx wigolo doctor` to confirm components are healthy; after restarting the client, the ten wigolo tools (search, fetch, crawl, extract, cache, find_similar, research, agent, diff, watch) should appear in the tool list, and a web-search prompt should return cited results with no API key.

Troubleshooting

  1. Run npx wigolo doctor to inspect component health.
  2. Retry failed downloads with wigolo warmup --all, or with --browser, --embeddings, or --reranker.
  3. If the browser will not launch on Linux, run wigolo warmup --browser and install the reported system libraries.
  4. For native build errors or unusual Node versions, use Node 20, 22, or 24 LTS.
  5. Behind a proxy, set USE_PROXY=true and PROXY_URL; TLS-inspecting proxies may also require NODE_EXTRA_CA_CERTS.
  6. Use --no-warmup to defer downloads until first use.

Things to try

Once connected, you can ask your AI assistant things like:

  • Search the web for the latest PostgreSQL logical replication docs and give me cited excerpts
  • Fetch https://example.com and convert it to clean markdown
  • Crawl all pages under a docs site sitemap and extract the Article schema metadata
  • Show me exactly what changed on this page since your last visit

Tools 10

search read-only
Runs parallel multi-engine search with rank fusion, reranking, and explainable result scores.
fetch read-only
Loads a URL through a tiered HTTP and headless-browser router and returns cleaned Markdown, metadata, and links.
crawl read-only
Crawls multiple pages using BFS, DFS, sitemap, or map-only modes while respecting robots.txt and per-domain rate limits.
extract read-only
Extracts tables, metadata, JSON-LD, brand identity, named schemas, or custom JSON Schema data from a page.
cache read-only
Searches previously seen content using keyword or hybrid semantic retrieval and supports statistics, clearing, and change detection.
find_similar read-only
Finds pages similar to a URL or concept using keyword, semantic, and live-web fusion.
research read-only
Decomposes a question, expands subqueries, fetches sources, and produces a cited report or structured brief.
agent read-only
Runs an autonomous gather loop covering planning, search, fetching, extraction, and synthesis with a step log and time budget.
Show 2 more tools
diff read-only
Shows what changed on a page since the previous visit.
watch read-only
Rechecks pages on demand and can deliver changes to a webhook.

Use cases

Providing AI coding agents with local web search and fetching
Crawling websites and extracting structured data
Building local RAG, research, or autonomous information-gathering workflows
Integrating self-hosted agents through MCP, REST, or SDKs

Supported clients

Claude Code
Cursor
Codex
Gemini CLI
VS Code
Windsurf
Zed
Antigravity

Listed from the project's documentation, not tested by this site.

Overview

wigolo is a local MCP and REST web-intelligence server for AI agents. It provides search, fetching, crawling, structured extraction, caching, similarity search, research, autonomous gathering, and page change monitoring. Core tools work without API keys, while cache, embeddings, models, and configuration are stored by default under ~/.wigolo/. The project is in public beta, requires Node.js 20 or newer, and needs about 1.5 GB of free disk space.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision da59eee72ee5 Data synced 2026-10-19 Read the FMRS scoring method