Skip to main content
Open Source · Apache-2.0

searchio Search

Keyless web search, page reading, and product lookup for humans and AI agents. A three-tier fetch ladder climbs bot walls and JavaScript-only pages — no API keys, no empty results.

View on GitHub Install
CORE_CAPABILITIES

Built for Deterministic Search

The same fetch ladder whether you run it from a terminal, a Python script, an HTTP client, or an AI agent

Keyless by Default

Search and page reading need no API keys and cost nothing. Results are fused and ranked across multiple keyless providers — you get one answer, not five tabs.

Blocking-Resistant Ladder

Three fetch tiers: plain HTTP, TLS impersonation, then a Rust engine sidecar and Chromium for the challenge tier. Per-domain tiers are learned and cached.

Honest Output

Every answer lists providers used, failed, and skipped. A walled page returns 451, provider failure returns 502 or 503 — you never get an empty page.

INSTALL

One Engine, Three Interfaces

Install once. Use it from the CLI, from an MCP client, or over HTTP — the ladder and the cache are shared.

Command Line

One pip install. Search, read pages, look up products, and run the research swarm from your terminal.

Features

search, read, items, marketplace commands
searchio doctor verifies every fetch tier
Python 3.10+, works on Windows, macOS, Linux
Optional Rust engine sidecar for tier 2
pip install "searchio[mcp]"

MCP Server

Six tools any MCP client can call: search, read URLs, product and local listings, research, and health stats.

Features

search, read_url, search_items, search_local
research — multi-step swarm with citations
stats — per-domain tiers and block rates
stdio, or HTTP with an x402 payment gate
npx -y @ioserviceslabs/searchio-mcp

HTTP API

searchio serve exposes the whole engine over HTTP — drop-in backend for apps, agents, and shared team infrastructure.

Features

GET /search, /read, /items, /marketplace
POST /research for the agent swarm
Per-session cookies on /read
/providers, /stats, /healthz status endpoints
searchio serve --port 8080
MCP_INTEGRATION

Give Your Agent the Web

Connect searchio to Claude Desktop, Cursor, or any MCP client in one step — the launcher picks Docker when present, otherwise your local install.

Setup Instructions

1

Install with pip, Docker, or the npm launcher

2

Add the configuration to your MCP client

3

Restart the client to activate the connection

4

Ask the agent to search the web — real results mean it works

Agent Capabilities

Fused, ranked web search with intents
Read pages through bot walls, get markdown
Product listings merged across retailers
Research swarm with cited findings
claude_desktop_config.json
{
    "mcpServers": {
        "searchio": {
            "command": "npx",
            "args": [
                "-y",
                "@ioserviceslabs/searchio-mcp"
            ]
        }
    }
}

No Keys Required

Search and reading need no API keys and cost nothing. Only the research swarm calls a paid model — set one key and it works.

USE_CASES

Beyond Search

The same ladder that reads a page also researches a question and prices a product

Research Swarm

A multi-step swarm of workers plans, searches, reads, and cross-checks — then returns findings, each citing the pages the workers actually read.

Details

DeepSeek or Anthropic as the model
Citations on every finding
Budget-capped by design

Product Intelligence

One row per product, prices merged across retailers. Facebook Marketplace by city, browser-backed and anonymous — credentials never sent anywhere.

Details

items command merges listings
Marketplace with local pickup radius
Anonymous browser tier

Built to Be Hosted

The MCP server speaks the x402 payment shape — a billable tool returns 402 with an X-Payment challenge until the caller pays. Host it next to SwarmIO for credits.

Details

x402 payment gate shape
Per-session cookie isolation
Full-ladder Docker image, state in a volume
GET_STARTED

Get searchio

Open source under Apache-2.0. One command gives your terminal — and your agents — the readable web.

View on GitHub Install Guide