Make Your SaaS Product
AI-Agent Accessible
Google's WebMCP protocol lets AI agents discover your features, start trials, search your docs, and help users inside your product. Through a structured contract built into the browser.
SaaS platforms with WebMCP markup become discoverable and actionable by AI agents -- a new distribution channel shipping in Chrome 146.
10 seconds. No signup. Free.
What WebMCP Means for Your Product
Your SaaS exposes structured tool definitions -- names, schemas, execute functions -- that AI agents call directly. Every claim below maps to the spec.
Your Site Becomes an API
Publish a contract -- tool names, descriptions, input schemas -- that tells any AI agent exactly what your product can do. The agent calls typed functions, not guessing at buttons.
| 1 | navigator.modelContext.registerTool({ |
| 2 | name: "get_pricing", |
| 3 | description: "Get pricing plans...", |
| 4 | inputSchema: { |
| 5 | type: "object", |
| 6 | properties: { |
| 7 | plan_type: { type: "string" } |
| 8 | } |
| 9 | }, execute: async (params) => { |
| 10 | return { content: [...] } |
| 11 | } |
| 12 | }); |
Two APIs -- Pick Your Path
Imperative -- JavaScript. Full control over complex multi-step flows.
Declarative -- HTML attributes on existing forms. Zero JS required. Most SaaS products ship both.
| 1 | <form toolname="start_trial" |
| 2 | tooldescription="Start a free |
| 3 | trial of [Product]." |
| 4 | toolautosubmit |
| 5 | action="/api/trial"> |
| 6 | <input type="email" |
| 7 | toolparamdescription="Work email" /> |
| 8 | <button>Start Trial</button> |
| 9 | </form> |
3 attributes turn an existing form into an agent-callable tool.
A Web Standard, Not a Vendor SDK
WebMCP is a proposed web standard co-authored by Google and Microsoft, shipping in Chrome 146. The navigator.modelContext API lives in the browser runtime.
When other browsers adopt it, your tools work everywhere. One implementation, full coverage.
“WebMCP bridges the gap between web applications and AI agents by providing a contract for interaction. Rather than the agent guessing the purpose of a button, the website explicitly publishes its interface.”
-- WebMCP Specification, Section 1.1
One Protocol Across Seven Lifecycle Stages
Discovery through renewal -- with tool definitions you can ship today.
A prospect tells their AI assistant: Find me a project management tool with Gantt charts, under $20/user, that integrates with Slack.
Agent searches the web, reads SEO listicles, recommends based on marketing copy. Your product competes on reputation and search ranking.
Agent queries get_pricing and get_features on YOUR SITE directly. Structured data: actual prices, feature lists, integration support.
navigator.modelContext.registerTool({ name: "get_features", description: "Get detailed feature list...", inputSchema: { properties: { category: { type: "string" } } }, execute: async ({ category }) => { ... } });
IMPACT Product recommended based on ACTUAL capabilities, not marketing claims.
NOTE Use Competitive Benchmarking to see how often agents select YOUR tool vs competitors'.
The prospect's agent has found 3 potential products. It needs to compare them side by side.
Agent reads marketing pages and comparison blogs that competitors control. Your pricing gets filtered through someone else's narrative.
Agent calls get_pricing on all three products simultaneously, gets structured JSON, builds an objective comparison.
navigator.modelContext.registerTool({ name: "get_pricing", inputSchema: { properties: { billing_cycle: { enum: ["monthly", "annual"] }, team_size: { type: "integer", min: 1 } } }, execute: async (params) => { ... } });
IMPACT You control the pricing narrative. Agent gets YOUR numbers directly.
NOTE team_size enables personalized evaluation at scale — something your pricing page can't do.
The prospect is convinced. They tell their agent: Start a free trial of [Product].
User finds signup page, fills form, verifies email, waits for onboarding. The agent can't help.
Existing trial form becomes agent-callable with 3 HTML attributes. Zero-friction trial in conversation flow.
<form toolname="start_trial" tooldescription="Start a free 14-day trial." toolautosubmit action="/api/trial"> <input type="email" toolparamdescription="Work email" /> </form>
IMPACT Trial starts in conversation flow. Zero friction. No landing page bounce.
NOTE Form Scanner detects your existing form and generates attributes automatically.
User just started trial: Set up my workspace with 3 channels: Engineering, Design, Marketing.
User navigates settings, watches tutorials, reads docs. Many abandon during this phase.
Agent calls setup_workspace directly. Workspace created with channels in one invocation.
navigator.modelContext.registerTool({ name: "setup_workspace", inputSchema: { properties: { workspace_name: { type: "string" }, channels: { type: "array" } } }, annotations: { readOnlyHint: false }, execute: async (params) => { ... } });
IMPACT Onboarding completion increases when agents perform setup actions directly.
NOTE Use provideContext() and dynamic tool registration for SPA state management.
Active user multitasking: Create a new project called Q3 Launch with a deadline of September 30.
User switches contexts, navigates menus, fills creation forms. Context-switching costs productivity.
Agent calls create_project directly. User stays in conversation flow. Product becomes part of AI workflow.
navigator.modelContext.registerTool({ name: "create_project", inputSchema: { properties: { name: { type: "string" }, deadline: { type: "string", format: "date" } }, required: ["name"] }, execute: async (params) => { ... } });
IMPACT Stickiness increases when product becomes part of user's AI-assisted workflow.
NOTE SubmitEvent includes agentInvoked boolean for analytics.
A user asks their agent a product question. Without WebMCP, the agent may hallucinate.
Agent searches the web, finds outdated docs, gives incorrect information. Or user submits a ticket.
Agent calls search_docs, which queries your ACTUAL documentation and returns the correct answer.
navigator.modelContext.registerTool({ name: "search_docs", inputSchema: { properties: { query: { type: "string" }, section: { enum: ["user-guide", "api-ref"] } } }, annotations: { readOnlyHint: true }, execute: async (params) => { ... } });
IMPACT Docs become a self-service product surface. Agents can answer user questions from your actual documentation.
NOTE Prompt Coverage Testing validates agents route questions correctly across phrasings.
User says: Upgrade our account to Pro for 25 users. This involves payment — requires confirmation.
User navigates billing, selects plan, enters payment, confirms. Multi-step, easy to abandon.
Agent calls upgrade_plan, calculates quote, triggers browser confirmation before processing.
navigator.modelContext.registerTool({ name: "upgrade_plan", annotations: { destructiveHint: true }, execute: async ({ target_plan, seats }) => { const confirmed = await requestUserInteraction({ title: "Confirm Upgrade", message: `$${quote.price}/mo` }); if (!confirmed) return { ... }; ... } });
IMPACT Expansion revenue becomes frictionless. Upgrading is a single sentence with confirmation.
NOTE requestUserInteraction() is mandatory for payment. The spec warns against financial actions without it.
When an Agent Evaluates You and Two Competitors Simultaneously
A VP of Engineering asks their agent: “Find me a CI/CD platform that supports monorepos, has security scanning, under $500/month for 20 users.”
Platform A didn't just implement WebMCP. They wrote precise tool descriptions with well-typed schemas and reliable execution. Agent selection rates correlate with tool definition quality, not just presence.
Tool quality compounds. Early optimization pays dividends in agent selection rates.
Where WebMCP Fits in Your Stack
Four zones. Two APIs. One protocol. Each maps to a different part of your product.
Marketing Site
Declarative (HTML)Low1-3 hoursAdd toolname, tooldescription, and toolparamdescription attributes to existing forms. Form Scanner auto-detects and generates attributes.
Zero to Score 91 in Four Weeks
A hypothetical walkthrough for a B2B SaaS product -- illustrating realistic timelines, effort, and score progression.
Quick Wins — Marketing Site
Added declarative WebMCP attributes to 3 forms: trial signup, demo request, pricing calculator.
3 hours · 12 lines of HTML attributes · No JavaScript changes
Documentation — search_docs
Wrapped existing Algolia-powered docs search API as a WebMCP tool. Optimized descriptions after prompt coverage testing.
4 hours · Prompt coverage: 87% → 94%
Product Integration
Implemented imperative API for core product actions with dynamic tool registration per view.
3 days · State-aware tools · Permission checking · provideContext() for SPA
Optimization
Tested across Gemini, GPT-4, and Claude. A/B tested tool descriptions for better routing.
Security Review & Launch
Ran security scanner, fixed permission gaps, added rate limiting. Final score: 91/100.
| Tool | Gemini | GPT-4 | Claude |
|---|---|---|---|
| create_project | 96% | 92% | 94% |
| search_docs | 94% | 91% | 93% |
| get_project_status | 88% | 72% | 85% |
| add_task | 91% | 89% | 78% |
The main engineering effort is writing clear tool descriptions -- and prompt coverage testing tells you when they route correctly across models. Existing forms, APIs, and docs search gain a new interface layer.
Your Agent Readiness Score
A number from 0-100 measuring how well your SaaS works with AI agents. Five weighted dimensions, inspired by Lighthouse.
Implementation + Security + Best Practices. Run from terminal or browser extension.
npx webmcp-cli audit https://yoursite.comAdds LLM-based Prompt Coverage and live Reliability testing across Gemini, GPT, and Claude.
Three Threat Categories Your Tools Must Handle
The WebMCP spec includes security annotations for a reason. Your Agent Readiness Score weights security at 20%.
Tool Poisoning
[CRITICAL]A tool says it does one thing but does another. The description promises "search docs" but the execute function scrapes user data.
Your users trust that your WebMCP tools do what they say. A poisoned tool erodes trust in your entire product.
Compares tool descriptions against execution behavior. Flags mismatches between stated intent and actual API calls.
Over-Parameterization
[HIGH]A tool requests more data than it needs. A "search docs" tool asking for email, company name, and billing info.
Over-parameterized tools leak customer data. Agents may automatically provide data users wouldn't manually enter.
Analyzes input schemas for unnecessary parameters. Flags tools requesting data beyond their stated function.
Misrepresentation
[HIGH]A tool misrepresents capabilities. Claims to support features it doesn't, or hides limitations in the schema.
When agents discover misrepresentation, they deprioritize your tools. This is Agent SEO negative ranking.
Tests every tool description claim against actual behavior. Validates schema accuracy with real invocations.
Six Steps to Agent-Ready
Start with a free CLI assessment. Ship your first agent-callable form in under an hour.
Assess — Free CLI
5 minBaselineRun the free Agent Readiness audit on your site. Get your current Quick Score, detected forms, and industry comparison.
Quick Wins — Marketing Forms
1-3 hours+15-25 ptsAdd declarative WebMCP attributes to existing marketing forms. Form Scanner detects your forms and generates attributes automatically.
Documentation — search_docs
4-8 hours+10-20 ptsExpose your documentation to agents. Wrap your docs search API (Algolia, Meilisearch, custom) as a WebMCP tool.
Product Core — In-App Actions
2-4 days+15-25 ptsUse the imperative API for core product actions. Start with read-only tools before write tools. Add provideContext() for SPA state management.
Test & Optimize
3-5 days+5-15 ptsRun Prompt Coverage Testing across models. Optimize tool descriptions using A/B testing. Fix security issues.
Launch & Monitor
OngoingMaintainDeploy analytics, add your badge, list on Discovery Hub, and set up continuous monitoring.
The AI Co-Pilot Option
The AI Co-Pilot crawls your site, identifies tool opportunities, generates definitions, and produces production-ready code. Turns a 4-week implementation into a 4-day review.
Start with AI Co-Pilot10 seconds. No signup required.
Your SaaS in Every
AI Agent Conversation
WebMCP shipped February 9, 2026 in Chrome 146. AI agents can already call structured tools on websites that implement the protocol.
Products with WebMCP become discoverable when agents evaluate options. Your Agent Readiness Score measures how well your product surfaces in those evaluations.
Run the free CLI audit to see where you stand today.
WebMCP is an open standard -- your implementation works across any browser that adopts it.