---

Abstract

Zero-API-Key Web Search is a local-first, MCP-native search and evidence-verification toolkit for AI agents. It provides live web search, LLM-optimized context extraction, claim verification with weighted evidence scoring, and citation-ready evidence reports — all without requiring an API key by default. The verification model (evidence-aware-heuristic-v3) classifies sources as supporting, conflicting, or neutral using keyword overlap, domain-quality heuristics, freshness, and optional page-aware rescoring. This project does not perform fact-level proof or logical entailment; it is a signal amplifier for agent grounding decisions.

Introduction

AI agents that rely on raw search retrieval often produce ungrounded or confidently wrong outputs because search results provide links, not evidence. Zero-API-Key Web Search bridges this gap by layering verification and source-quality scoring on top of live search. The research question it addresses is: Can heuristic evidence scoring and source-quality weighting reduce ungrounded agent outputs compared to raw search retrieval?

Why Agents Use It

A single install gives your agent live search, page reading, claim verification, and citation-ready context without requiring an API key by default.

Agent job	Command	What the agent gets
Ground an answer	zero-context "FastAPI lifespan docs"	compact Markdown context with citations
Verify a claim	zero-verify "Python 3.13 is the latest stable release"	supported / contested / likely false verdict
Build an evidence report	zero-report "AI regulation news"	rationale, source digest, warnings, next steps
Serve an MCP client	zero-mcp	search_web, llm_context, browse_page, verify_claim, evidence_report

Quick start

pip install zero-api-key-web-search

# Search the web
zero-search "Python 3.13 release" --json

# Inspect provider options
zero-search providers

# Build citation-ready LLM context
zero-context "Python 3.13 stable release" --goggles docs-first

# Read a page
zero-browse "https://docs.python.org/3/whatsnew/" --json

# Verify a claim
zero-verify "Python 3.13 is the latest stable release" --deep --json

# Full evidence report
zero-report "Python 3.13 stable release" \
  --claim "Python 3.13 is the latest stable release" --deep --json

Legacy CLI aliases (search-web, browse-page, verify-claim, evidence-report) also work.

Node-based agent runtimes can use the npm wrapper. The wrapper source is included in this repository; see docs/npm-package.md.

The 30-Second Pitch

• Zero-key default: useful immediately for local agents, evals, demos, and prototypes.
• MCP-native: works as a reusable tool server for Claude Code, Cursor, Copilot-style clients, Codex, Gemini, OpenClaw, and other MCP-compatible runtimes.
• LLM-context first: zero-context returns context a model can actually use, not just a pile of links.
• Evidence-aware: zero-verify and zero-report preserve support, conflict, source quality, freshness, and domain diversity (within the heuristic boundary described in docs/trust-model.md).
• Provider-aware: start free with ddgs, add self-hosted searxng, or opt into Bright Data for production reliability and geo-targeting.

Why this over a plain search wrapper?

Feature	Plain search	Zero-API-Key Web Search
Live search results	✅	✅
News / images / videos / books	❌	✅
Region & time filtering	❌	✅
Full-page text extraction	❌	✅
Claim verification with evidence scores	❌	✅
Supporting vs. conflicting evidence	❌	✅
Citation-ready evidence reports	❌	✅
Dual-provider cross-validation	❌	✅
API key required	Often	Never by default (Default provider is DuckDuckGo; no key needed. Production providers require configuration.)
Cost	Sometimes	Free by default

Compare the Shape

Project shape	Best at	Tradeoff
Plain search wrapper	returning links quickly	leaves grounding, citation shaping, and conflict handling to the agent
Hosted search API	managed reliability and scale	usually requires an account/key from the first request
Zero-API-Key Web Search	local agent search infrastructure with optional production providers	default results depend on free upstreams unless you add SearXNG or Bright Data

This project is not trying to be a hosted search engine. It is the missing search/evidence layer inside agent runtimes.

MCP server

Works with Claude Code, Cursor, Copilot, and any MCP-compatible agent:

{
  "mcpServers": {
    "zero-api-key-web-search": {
      "command": "zero-mcp"
    }
  }
}

For npm/npx-based MCP launchers after npm publication:

{
  "mcpServers": {
    "zero-api-key-web-search": {
      "command": "npx",
      "args": ["zero-api-key-web-search", "zero-mcp"]
    }
  }
}

Six tools exposed: list_providers, search_web, llm_context, browse_page, verify_claim, evidence_report.

Platform support

Platform	Status	Entry point
CLI	Ready	zero-search, zero-context, zero-browse, zero-verify, zero-report
MCP	Ready	zero-mcp
Claude Code	Ready	.claude/skills/zero-api-key-web-search/SKILL.md
Gemini	Ready	GEMINI.md + .gemini/SKILL.md
Cursor	Ready	.cursor/rules/zero-api-key-web-search.md
Copilot	Ready	.github/copilot/instructions.md
Codex	Ready	.codex/SKILL.md
Continue	Ready	.continue/skills/zero-api-key-web-search/SKILL.md
Manus	Ready	Root SKILL.md + docs/manus.md
Kiro	Ready	.kiro/steering/zero-api-key-web-search.md
OpenClaw	Ready	zero_api_key_web_search/skills/SKILL.md

How verification works

zero-verify uses the evidence-aware heuristic v3 model:

1. Search for the claim across available providers
2. Score each source on keyword overlap, source quality, freshness
3. Classify as supporting, conflicting, or neutral
4. Optionally fetch top pages for deeper page-aware analysis
5. Render a verdict with confidence and evidence breakdown

Verdict	Meaning
supported	Strong evidence, low conflict
likely_supported	Leans positive, not decisive
contested	Support and conflict both meaningful
likely_false	Conflict strong, support weak
insufficient_evidence	Too weak for any firmer verdict

This is a heuristic evidence classifier, not a proof engine. See docs/trust-model.md for details and limitations, docs/verification-model.md for signal definitions, and docs/benchmarks.md for regression results.

Free dual-provider setup

Default install uses DuckDuckGo. For stronger cross-validated evidence, add a free self-hosted SearXNG:

./scripts/start-searxng.sh
export ZERO_SEARCH_SEARXNG_URL="http://127.0.0.1:8080"
./scripts/validate-free-path.sh

Or with Docker Compose:

cp .env.searxng.example .env
docker compose -f docker-compose.searxng.yml up -d

Full guide: docs/searxng-self-hosted.md.

Agent search controls

Provider profiles make backend choice explicit:

Profile	Providers	Best for
free	ddgs	zero-setup local use
free-verified	ddgs, searxng	free cross-validation
production	brightdata	production reliability and geo-targeting
max-evidence	ddgs, searxng, brightdata	maximum provider diversity

zero-search "FastAPI lifespan docs" --profile free-verified --goggles docs-first
zero-context "FastAPI lifespan docs" --profile free --goggles docs-first
zero-report "AI regulation news" --profile production --json

Built-in Goggles-lite presets:

Goggles	Effect
docs-first	boosts docs, API, support, release-note, and official-looking sources
research	boosts academic, institutional, paper, and study-oriented sources
news-balanced	boosts reporting/analysis signals and demotes low-context aggregators

You can also pass a JSON file to --goggles with boost_domains, block_domains, demote_domains, and boost_title_terms.

Full guide: docs/agent-search-controls.md.

More agent integration material:

• Agent Developer Guide
• Demo Transcript
• Launch Kit

Optional Bright Data provider

The default path stays free and zero-key. For production agents that need higher reliability, structured SERP data, geo-targeted results, or stronger cross-provider verification, enable the optional Bright Data provider.

export ZERO_SEARCH_BRIGHTDATA_API_KEY="..."
# Optional if your Bright Data SERP zone is not named web_search:
export ZERO_SEARCH_BRIGHTDATA_ZONE="web_search"

zero-search providers
zero-search "AI regulation news" --provider brightdata --type news --region us-en --json
zero-report "Tesla Q1 2026 deliveries" \
  --claim "Tesla deliveries increased year over year" \
  --provider ddgs --provider brightdata --deep --json

New Bright Data users can sign up here: https://get.brightdata.com/h21j9xz4uxgd.

Provider	Best for
ddgs	zero-setup local search
searxng	free self-hosted cross-validation
brightdata	production-grade, geo-targeted, structured SERP evidence

Evidence report example

{
  "verdict": "contested",
  "confidence": "MEDIUM",
  "executive_summary": "Evidence is split...",
  "verdict_rationale": ["Source A supports...", "Source B contradicts..."],
  "coverage_warnings": ["Single-provider evidence path."],
  "source_digest": [
    {"title": "...", "url": "...", "classification": "supporting", "evidence_strength": 3}
  ],
  "next_steps": ["Add a second provider for cross-validation."]
}

Architecture

zero_api_key_web_search/
  core.py              # UltimateSearcher — search, verify, report engine
  browse_page.py       # Readability-style page text extraction
  mcp_server.py        # MCP server (6 tools)
  transport.py         # SSL/TLS helpers
  search_web.py        # CLI: zero-search
  context.py           # CLI: zero-context
  browse_page.py       # CLI: zero-browse
  verify_claim.py      # CLI: zero-verify
  evidence_report.py   # CLI: zero-report
  providers/
    base.py            # SearchProvider protocol (sync + async)
    ddgs.py            # DuckDuckGo provider
    searxng.py         # SearXNG provider
    brightdata.py      # Optional Bright Data SERP provider
  skills/
    SKILL.md           # Bundled OpenClaw skill

Key engineering features:

• Circuit breaker: Trips after 3 consecutive provider failures, auto-resets after 60s
• Async support: asearch() for concurrent provider calls via asyncio.gather
• Structured logging: Configurable logging at search/verify/report entry points
• Readability heuristic: Semantic HTML5 + ARIA roles + text density scoring
• Baseline comparison: Majority-vote and keyword-count baselines in reports
• Sub-claim decomposition: Targeted sub-queries for independent evidence gathering

Installation

pip install zero-api-key-web-search

Python 3.10+ required. No API keys, no accounts, no configuration needed.

Optional npm wrapper source:

npm pack --dry-run

Development

pip install -e ".[dev]"
python -m pytest tests/ -q           # 98 tests
ruff check zero_api_key_web_search/ tests/
pyright zero_api_key_web_search/     # 0 errors
coverage report --fail-under=80       # 85% coverage

Verification for ecosystem reviewers

1. zero-report "Python 3.13 stable release" --claim "Python 3.13 is the latest stable release" --deep --json
2. docs/ecosystem-readiness.md
3. docs/gemini-submission-checklist.md
4. docs/trust-model.md

License

MIT License.