mcp-omnisearch MCP — Use Cases, Install & Live Demo

Why use it

Key features

4 tools cover search, AI-search, GitHub, and extraction — no tool-sprawl
Bring-your-own-keys per provider; use only what you pay for
Tavily, Brave, Kagi, Exa for search; Kagi FastGPT, Exa, Linkup for AI search
Firecrawl self-hosted support via FIRECRAWL_BASE_URL
Docker + OpenAPI (MCPO) available for remote deployments

Live Demo

What it looks like in practice

omnisearch.replay ▶ ready

0/0

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "omnisearch": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-omnisearch"
      ],
      "_inferred": true
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "omnisearch": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-omnisearch"
      ],
      "_inferred": true
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "omnisearch": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-omnisearch"
      ],
      "_inferred": true
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "omnisearch": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-omnisearch"
      ],
      "_inferred": true
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "omnisearch",
      "command": "npx",
      "args": [
        "-y",
        "mcp-omnisearch"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "omnisearch": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp-omnisearch"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add omnisearch -- npx -y mcp-omnisearch

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use mcp-omnisearch

Research a topic with fallback across multiple search engines

👤 Researchers, analysts, anyone who's been burned by a single engine's bias ⏱ ~15 min beginner

When to use: A fresh topic where one engine's top-10 often differs from another's.

Prerequisites

At least one search API key (Tavily free tier is easiest) — tavily.com/signup — free tier is 1000 req/month

Flow

Run a search with the cheapest provider first

Search Tavily for 'latest MCP protocol changes 2026'. Summarize the top 5 results.✓ Copied

→ Summary with citations
Cross-check on a different provider

Now search the same query on Brave. Which results show up in both? Which are unique?✓ Copied

→ Overlap analysis
Extract full content for the most promising result

Use web_extract on the top result and give me the actual useful text, not nav boilerplate.✓ Copied

→ Clean markdown of the article

Outcome: A cross-validated research answer with confidence boosted by multi-engine agreement.

Pitfalls

Per-provider quotas vary — Tavily is generous, Kagi costs per query — Route exploratory searches to Tavily, save Kagi for high-signal 'give me the answer' queries via FastGPT
web_extract fails on JS-heavy sites — Use Firecrawl (JS-rendering) for SPAs; Tavily extract for static pages

Combine with: notion

Get a cited quick answer with AI search

👤 Anyone who wants 'what's the answer' not 'here are 10 blue links' ⏱ ~5 min beginner

When to use: A specific factual question where you want reasoning with sources.

Prerequisites

At least one AI-search key: KAGI_API_KEY, EXA_API_KEY, or LINKUP_API_KEY — Kagi FastGPT is the cheapest per query

Flow

Ask via ai_search

Use ai_search (Kagi FastGPT) for: what's the default rate limit for the GitHub Models API as of Q1 2026?✓ Copied

→ Direct answer with citations
Verify the citation

Open the top citation with web_extract and confirm the claim.✓ Copied

→ Quoted source text

Outcome: A verified quick answer, cheaper and faster than a free-search round-trip.

Pitfalls

AI search can confidently misquote sources — Always web_extract the top citation for claims that matter

Search public GitHub for reference implementations

👤 Devs evaluating libraries or looking for code patterns ⏱ ~15 min intermediate

When to use: 'Has anyone solved X in Rust?' type questions.

Prerequisites

GITHUB_API_KEY with no scopes (public only) — github.com/settings/tokens — classic token with empty scope list is fine

Flow

Search with code operators

github_search for fn main() language:rust path:src/ 'tokio::select!' — top 20 repos.✓ Copied

→ Repo list with file hits
Pull specific files worth reading

For the top 3 hits, use web_extract on the raw.githubusercontent URLs and show me the tokio::select! patterns they use.✓ Copied

→ Code excerpts with context

Outcome: A focused reading list of real-world implementations you can learn from.

Combine with: github

Combinations

Pair with other MCPs for X10 leverage

omnisearch + notion

Research a topic across engines, save distilled notes to Notion

Research 'post-quantum TLS 2026 state of play' across Tavily and Brave, extract the 3 best sources, and save a Notion page summarizing them.✓ Copied

omnisearch + github

Discover repos on GitHub public search, then do deep ops on the best one via the github MCP

github_search for rate-limiter libraries in Go, pick the top, then via the github MCP open its latest 3 issues.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
web_search	query: str, provider?: tavily\|brave\|kagi\|exa, options?: {site?, filetype?, time?, lang?}	Broad web search with operator control	1 provider API call
ai_search	query: str, provider?: kagi_fastgpt\|exa\|linkup	You want a direct cited answer, not links	1 AI-search call (costlier than regular search)
github_search	query: str, kind?: code\|repo\|user	Code/repo/user discovery with GitHub syntax	1 GitHub API call (30/min code search limit)
web_extract	url: str, provider?: firecrawl\|tavily\|kagi	Get clean article text from a URL	1 extractor call

Cost & Limits

What this costs to run

API quota: Per-provider. Tavily: 1000 req/mo free. Brave: 2000/mo free. Kagi: pay-per-call. Exa: free tier ~1000/mo.
Tokens per call: Search results: 500-1500 tokens. AI search with citations: 1-3k. Extracted articles: 2-8k.
Monetary: Free to start (Tavily, Brave, Exa free tiers); Kagi pay-per-call; Firecrawl self-host = free
Tip: Route: cheap broad search → Tavily/Brave; high-signal answers → Kagi FastGPT; extraction → self-hosted Firecrawl.

Security

Permissions, secrets, blast radius

Minimum scopes: GitHub token: empty scopes (public only)

Credential storage: Env vars per provider: TAVILY_API_KEY, BRAVE_API_KEY, etc.

Data egress: Queries to each configured provider's API; results back through

Never grant: GitHub token with repo write — web_search is public-only Self-hosted Firecrawl without auth exposed to public internet

Each provider's TOS applies to your usage — don't scrape with one, extract with another, and re-publish.
Providers log queries. Don't search for secrets or internal-only terms.

Troubleshooting

Common errors and fixes

Provider not configured

Set the env var for that provider, or pass provider= to route to a different one.

Verify: echo $TAVILY_API_KEY

Firecrawl extraction timeout

Heavy SPA; raise FIRECRAWL timeout, or fall back to tavily/kagi extract for static pages.

GitHub search returns 403

Unauthenticated code search is heavily rate-limited. Supply GITHUB_API_KEY even with empty scopes.

Verify: curl -H 'Authorization: Bearer $GITHUB_API_KEY' https://api.github.com/rate_limit

Alternatives

mcp-omnisearch vs others

Alternative	When to use it instead	Tradeoff
brave-search MCP	You only need one engine and want a minimal config	No AI search, no multi-provider redundancy
searxng MCP	You want fully self-hosted metasearch with no API keys	You host the SearXNG instance
kindly-web-search	You want fuller content retrieval baked in (full StackOverflow threads, GitHub issues)	Smaller provider list

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills