WEBMCP PLATFORM · CHROME 146+
Scan, Build, Test, and Ship Agent-Ready Tools
WebMCP launched in Chrome two weeks ago. Web-MCP is a platform built for the new standard — from your first registerTool() call to production monitoring and optimization.
60-second site scan
3-model prompt testing
0-to-100 readiness score
Real-time agent analytics
Free tier available · No credit card required · Works with Chrome 146+
Without Web-MCP
Read 40-page spec
Learn JSON Schema
Write registerTool() by hand
Test in DevTools manually
40-60 hrs/site
With Web-MCP
Paste URL
AI scans your site
Review tool definitions
Export & ship
< 1 hour/site
Why This Matters Now
WebMCP is a new web standard from Google and Microsoft that lets websites expose structured tools for AI agents — replacing screen-scraping with reliable, schema-driven interaction.
LIVEIt launched in Chrome 146 on February 9, 2026.
Every website with forms, search, checkout, or interactive features will need WebMCP tools. The specification is finalized and shipping in production browsers.
Web-MCP helps you implement, test, and monitor your WebMCP tools.
CHOOSE YOUR PATH
I'm a...
CREATE
From Zero to Agent-Ready in Under an Hour
WebMCP implementation that used to take weeks of manual work, handled by AI-assisted tooling.
Without Web-MCP
- Read the 40-page WebMCP spec
- Learn JSON Schema draft-07
- Write registerTool() calls by hand
- Debug agentInvoked event handling
- Test in DevTools manually
40-60 hours per site
With Web-MCP
- Paste your URL
- AI Co-Pilot scans your site
- Review generated tool definitions
- Export production code
- Ship.
Under 1 hour per site
8.7
“Your AI Implementation Partner”
AI Co-Pilot
Point it at any URL. Get a complete WebMCP implementation plan. The Co-Pilot crawls your website, identifies every form and interactive element, and generates a complete implementation.
- Full registerTool() calls with proper inputSchema and MCP annotations
- Declarative HTML attributes for existing <form> elements
- Security audit flagging injection risks and missing safeguards
- Prioritized roadmap: high-value tools first
- Framework-specific code exports with feature detection
“The Co-Pilot analyzes your site structure, generates tool definitions, reviews security posture, and produces an implementation roadmap — turning weeks of manual spec reading into an automated workflow.”
Ready
https://example-hotel.com
room-search5 params \u00b7 98%
contact-form3 params \u00b7 95%
newsletter-signup1 params \u00b7 87%
Readiness Score—
7.8
“Any URL → Agent-Ready Tools in 60 Seconds”
Form Scanner
Paste a URL. The scanner detects every <form>, <input>, <select>, and interactive element. For each one, it generates tool definitions with confidence scores.
- Suggested toolname derived from form action, ID, or button text
- tooldescription following WebMCP best-practice framing
- inputSchema built from input types with smart mapping
- Confidence score per suggestion
- Flags for hidden inputs and JS-driven elements
https://example-hotel.com/booking
Found 4 forms:
✓
Room Search
98% \u00b7 5 params
✓
Contact Form
95% \u00b7 3 params
!
Newsletter
87% \u00b7 1 params
⚙
Date Picker
JS-driven, needs imperative tool
TEST
Will Agents Actually Use Your Tools Correctly?
The answer matters more than anything else you'll build. We test it across every model, every phrasing, every edge case.
Without Web-MCP
- Open ChatGPT
- Paste tool definitions
- Try 50 phrasings manually
- Track results in a spreadsheet
- Repeat for each model
4-6 hours per tool, per model
With Web-MCP
- Click "Run Prompt Coverage"
- 200 prompts × 3 models = tested
- See routing accuracy per model
- Get specific fix suggestions
- Integrate into CI/CD pipeline
Automated in minutes
8.4
“Test Whether Agents Actually Route to Your Tools”
Prompt Coverage Testing
We send your tool definitions alongside 200+ natural language prompts to Gemini, GPT, and Claude simultaneously. See exactly which prompts work, which fail, and why.
- Tool routing accuracy — Did the model pick the right tool?
- Parameter extraction accuracy — Correct values from prompt?
- Hallucination rate — Did it invent parameters?
- Clarification behavior — Did it ask for missing fields?
- Tool disambiguation — Did it confuse similar tools?
“Automated multi-model prompt testing replaces the manual process of testing phrasings one by one. Run hundreds of test cases across models in minutes instead of weeks.”
Prompt Coverage Matrix
PromptGemGPTCla
"cheap flights to NYC"
✓
✓
✓
"vol pas cher pour NY"
✓
✗
✓
"fligth frum london"
✓
✓
~
"book the cheapest"
✗
✓
✓
"I need to get there"
~
✗
✓
Routing accuracy:94.2%Sample data
7.8
“Automated Agent Readiness on Every Deploy”
CI/CD Quality Gates
One YAML file. One CLI command. Agent readiness becomes an automated quality gate in your existing pipeline — like Lighthouse CI for WebMCP.
- Set minimum score thresholds
- Configure prompt coverage requirements
- Block on security issues
- Set monthly LLM cost budgets
web-mcp.tests.yml
min_score: 75
prompt_coverage:
models: [gemini-2.0, gpt-4o, claude-sonnet]
min_routing_accuracy: 0.90
security:
block_on: [injection-risk]Deploy: APPROVEDScore: 81/100
OPTIMIZE
Understand How Agents Choose Between Competing Tools
When an AI agent has access to your tools AND your competitor's tools, who wins? Now you can measure it.
Without Web-MCP
- Guess at description wording
- No visibility into agent preferences
- Competitor analysis: impossible
- A/B testing: manual and slow
Weeks of trial and error
With Web-MCP
- Data-driven description optimization
- Multi-model preference insights
- Real-time competitive benchmarking
- Statistical A/B testing
Insights in seconds
7.9
“One Number. 0-100. How Agent-Ready Is Your Site?”
Agent Readiness Score
Your Agent Readiness Score is a composite metric weighing Implementation Quality (25%), Prompt Coverage (25%), Security Posture (20%), Reliability (15%), and Best Practices (15%).
- Implementation Quality — Schema validity, description quality
- Prompt Coverage — Agent selection and parameterization reliability
- Security Posture — Injection resistance, proper safeguards
- Reliability — Error handling, graceful failure
- Best Practices — Naming conventions, MCP annotations
“A single, transparent metric that maps directly to the WebMCP specification. Every check is documented, every weight is disclosed.”
Agent Readiness Score
0
Implementation
78
Prompt Coverage
91
Security
72
Reliability
84
Best Practices
88
Score: 81 / 100Sample data
8.3
“See How Your Tools Compare to Competitors”
Competitive Benchmarking
We present anonymized tool definitions from sites in your category to LLMs alongside realistic prompts. See your selection rate vs. other implementations.
- Selection rate — How often agents choose your tool
- Why competitors win — Specific description advantages
- Parameter accuracy comparison — Value extraction quality
- Agent preference patterns — Which models prefer which tools
“Data-driven competitive intelligence for the agent era. Understand exactly why agents prefer one tool description over another.”
Agent Selection Rate
Industry Leader48%
Your Tool (after)51%
Competitor B29%
Competitor C22%
Example: Specific parameter descriptions (e.g. “IATA airport codes”) can improve agent parameter accuracy.
PROTECT
Security Mapped to the WebMCP Specification
Every feature maps directly to the threat vectors documented in the official WebMCP Security specification.
Without Web-MCP
- Manual security review
- No injection detection
- Compliance docs: 4-8 weeks
- No runtime protection
Months of security work
With Web-MCP
- Automated vulnerability scanning
- Real-time injection detection
- Auto-generated compliance reports
- < 5ms runtime firewall
Protected from day one
Enterprise consideration: Security and compliance features are often the prerequisite for enterprise adoption. The WebMCP Security specification documents specific threat vectors that these tools scan for, including tool poisoning, output injection, and over-parameterization.
9.0
“Documentation to Help Your Legal Team Evaluate WebMCP”
Compliance Report Generator
Auto-generated compliance documentation mapped to GDPR, HIPAA, PCI-DSS, and accessibility frameworks. Data inventory, consent requirements, and flow diagrams.
- GDPR — Data inventory, consent, data minimization analysis
- HIPAA — Safeguard verification for healthcare data
- PCI-DSS — Payment parameter isolation verification
- Accessibility — Agent pathway accessibility assessment
“Auto-generated compliance documentation reduces the time legal teams need to evaluate WebMCP implementations, covering the regulatory frameworks they care about.”
Compliance ReportGENERATED
GDPR
94%12 items
HIPAA
88%8 items
PCI-DSS
91%6 items
A11y
85%4 items
Output: Structured compliance report
MEASURE
See What Agents Actually Do on Your Site
Production analytics that answer the two questions every stakeholder asks: "Is this working?" and "Is this making us money?"
Without Web-MCP
- No agent traffic visibility
- Unknown conversion attribution
- Error diagnosis: guesswork
- No journey mapping
Flying blind
With Web-MCP
- Agent traffic detection and classification
- Full revenue attribution
- Intelligent error diagnostics
- Complete journey visualization
Full visibility
8.4
“Track What AI Agents Are Worth to Your Business”
Revenue Attribution
Track every agent-driven interaction from first tool invocation to completed conversion. See agent-driven revenue, conversion rates, and top-performing tools.
- Agent-driven revenue this month
- % of total conversions from AI agents
- Agent vs. human conversion rate comparison
- Per-tool revenue attribution
“Production-grade analytics that connect agent interactions to business outcomes, giving stakeholders concrete data on WebMCP ROI.”
Agent Revenue (Demo)+340%
$0
Agent %
8.3%
vs. Human
71% vs 54%
Top tools
1.
search-flights$6,2002.
book-hotel$3,8003.
search-car-rental$2,4007.6
“Agent vs. Human — Know Who's Using Your Site”
Traffic Intelligence
Using SubmitEvent.agentInvoked plus behavioral heuristics, we separate agent traffic with high accuracy. See agent percentage, model breakdown, and trends.
- Agent traffic percentage
- Agent-by-model breakdown (Gemini, GPT, Claude)
- Trend analysis — agent adoption velocity
- Geographic and temporal patterns
Traffic SplitLast 30 days
AI Agents8.3%
Human91.7%
Trend:+340% MoM
GROW
Get Discovered. Get Chosen. Get Ahead.
The WebMCP spec has no built-in discovery mechanism. We're building it.
Without Web-MCP
- No discovery mechanism exists
- Agents can't find your tools
- Zero network effects
- Isolated implementation
Invisible to agents
With Web-MCP
- Listed in the Discovery Hub
- Verified certification badge
- Template marketplace access
- Growing network effects
Discoverable globally
Why a Discovery Registry
The WebMCP specification defines how tools work but not how agents find them. The Discovery Hub fills this gap as a public, searchable registry where sites list their tools and agents can look them up. The more sites that register, the more useful the registry becomes for agent platforms.
8.1
“A Registry That Connects AI Agents to Your Website”
Discovery Hub
The WebMCP spec has no discovery mechanism. The Discovery Hub is a public, searchable registry where websites list tools and AI agents discover them.
- Register your tools publicly
- Tools verified and categorized
- Verified listing with your Agent Readiness Score
- Opt into competitive benchmarking
- Drive agent traffic to your site
“The WebMCP spec defines how agents interact with tools, but not how agents find them. The Discovery Hub fills that gap — a canonical registry connecting agents to websites.”
Network Growth
100Proof of concept
1KPlatforms start querying
10KDefinitive registry
100KInfrastructure layer
§07—PROTOCOL COVERAGE
Protocol Native
Deep WebMCP Protocol Support
Web-MCP tools map directly to the WebMCP API surface — registerTool(), provideContext(), agentInvoked events, and MCP annotations.
WebMCP API / Concept
Web-MCP Feature
Implementation
API
navigator.modelContext.registerTool()FeatureTool Studio, Code Gen
ImplementationGenerate spec-compliant tool registrations
API
navigator.modelContext.provideContext()FeatureSPA State Manager
ImplementationManage dynamic tool availability
API
navigator.modelContext.clearContext()FeatureSPA State Manager
ImplementationGenerate cleanup on navigation
API
SubmitEvent.agentInvokedFeatureAnalytics SDK
ImplementationDistinguish agent vs human traffic
API
SubmitEvent.respondWith()FeatureCode Generation
ImplementationGenerate declarative form handling
API
toolactivated / toolcancelFeatureTest Console
ImplementationMonitor agent interaction lifecycle
API
requestUserInteraction()FeatureSecurity Scanner
ImplementationVerify destructive action consent
API
toolname / tooldescriptionFeatureForm Scanner
ImplementationGenerate declarative HTML annotations
API
JSON Schema inputSchemaFeatureAll authoring tools
ImplementationFull draft-07 with WebMCP best practices
API
MCP AnnotationsFeatureVisual Builder
ImplementationreadOnlyHint, destructiveHint, idempotentHint
Each feature maps to a specific WebMCP API call or event: registerTool(), provideContext(), agentInvoked, and the full MCP annotation set.
INTEGRATIONS
Integrates Into Your Existing Workflow
CLI for your terminal. Extension for your browser. Plugins for your CMS. SDK for your code.
Development
CLI Tool
Audit, lint, score from terminal
Browser Extension
Inspect WebMCP tools on any site
VS Code Extension
SoonInline validation in editor
CI/CD
GitHub Actions
Quality gates in your workflow
GitLab CI
Same CLI, any CI environment
CMS
Shopify Plugin
One-click install for stores
WordPress Plugin
WooCommerce + form support
Webflow
SoonNative Webflow integration
Framework
Next.js SDK
React hooks for WebMCP
Analytics SDK
< 5KB production monitoring
Firewall SDK
Runtime protection
API
REST API
Programmatic access
Running Shopify or WordPress?
Install our plugin. It auto-detects your forms, generates WebMCP tools, and starts tracking agent traffic — no code changes required.
§08—GET STARTED
WebMCP Is Live. Get Started.
WebMCP is shipping in Chrome. Scan your site, generate tool definitions, test across models, and monitor agent traffic. Try it free.
Test Your Site
Paste your URL and see your agent readiness in 60 seconds.
Get Your Score
Score + lint + recommendations in your terminal.
$ npx @anthropic-ai/web-mcp auditTalk to Sales
Custom pricing, compliance reports, dedicated support.
Built for the WebMCP standard
Works with Chrome 146+
Free tier available
See these features in action: Browse ready-to-use industry templates
Questions everyone's asking
Straight answers for business owners, developers, and teams evaluating Web-MCP
Web-MCP (Model Context Protocol) is a standard that lets AI agents interact with your website programmatically — like an API that AI agents can discover automatically. As AI-powered browsing grows, sites that implement structured agent interfaces will be easier for AI to understand and recommend. Getting ahead of this shift gives your business an advantage as the ecosystem matures.
Google hasn't confirmed Web-MCP as a direct ranking factor. However, Google has historically promoted web standards that improve machine readability — mobile-friendly, HTTPS, and Core Web Vitals all followed this pattern. Implementing Web-MCP ensures your site is prepared if structured agent data becomes a signal, and it helps AI agents understand your content regardless.
Web-MCP offers a free compliance scanner to audit any site instantly. For full implementation, our Pro plan starts at $49/month and includes schema generation, endpoint building, and monitoring. The implementation process is designed to be lightweight — no backend team or code changes required for most sites.
No. Web-MCP works as an additive layer on top of your existing site — no redesign, no migration, no downtime. We generate MCP-compatible endpoint definitions that sit alongside your current pages. Your human visitors see nothing different. AI agents see structured tools they can interact with.
As AI agents become a more common way for users to browse and transact, sites without structured agent interfaces may be harder for AI to recommend or interact with. Web-MCP gives you control over what agents can do — you define the tools, set the parameters, and track every interaction.
Any agent that supports the Model Context Protocol. The protocol is open and backed by major AI platforms, so adoption is growing across the ecosystem. As more AI assistants support MCP, your endpoints become accessible to a wider range of agents automatically.
Security is built into the protocol. You explicitly define which tools are available, what parameters they accept, and what permissions they require. Web-MCP never stores your raw content. All communications are encrypted, and you control exactly what AI agents can access. We also provide monitoring for agent interactions.