Check a website
Inspect public source. Get the finding, evidence and next step together.
Start a checkDeveloper tools for agent-facing websites
Check your site.
Fix the gaps.
Catch regressions.
Find source-level issues in your public page, understand the evidence, and repeat checks from your terminal or CI.
Get SDK + CI toolkit Or check a public page nowNo account needed for a public-source check. WebMCP is experimental; a scan does not execute your tools or prove agent task success.
Start with your page
What can we find in your page source?
The scan fetches only the public page you provide. It does not sign in, execute the target's JavaScript, or use your browser cookies. We retain the normalized public origin and path plus the outcome for 90 days, but never URL credentials, queries, fragments, goals, page markup, IP addresses, or user agents. Do not paste signed or secret-bearing URLs.
Site tools: graceful fallback active
Audit tool contracts
Inspect supplied tool definitions for structural issues before runtime testing.
Open contract auditAdd release checks
Save a reviewed baseline. Compare compatible source findings after changes.
Use the developer toolkitSame task · same app · two action surfaces
AI agents shouldn't have to guess where to click.
WebMCP lets a website expose actions like search, compare, and add to cart as typed tools. isWebMCP shows what changes—and checks whether the result can actually be verified.
Controlled comparison
What changes when the page exposes intent?
Synthetic headset task · find, compare, cart, then stop
The agent must operate the interface
Every step depends on discovering and interpreting UI state.
- 01Enter a maximum price
- 02Choose a minimum rating
- 03Enter minimum battery life
- 04Select noise canceling
- 05Apply the filters
- 06Inspect the result cards
- 07Select two eligible products
- 08Open the comparison
- 09Add the best match to the cart
- 10Inspect the cart to verify the result
Aurora Q45 selected
Same visible cart result through either path.
7/7
task checks pass
Checkout was not attempted. The safety boundary held.
01
The page names the action
An agent can discover “search products” instead of reverse-engineering fields, buttons, and result cards.
02
The schema checks the input
Price, rating, battery, and feature constraints arrive as typed fields instead of ambiguous prose.
03
The result can be verified
Stable product IDs and visible cart state make success testable while checkout stays out of bounds.
The report is the product
One URL. A ledger of what is known.
The source scan is intentionally conservative. It analyzes public source, maps candidate actions, and ranks next steps—while refusing to invent runtime proof it cannot observe.
- Every finding links back to exact, sanitized evidence.
- Unknown runtime quality stays unknown—not zero.
- Source, runtime, imported, measured, and inferred evidence never blur together.
Synthetic fixture · not a website observation
demo.iswebmcp.com/store
6
Source categories
Every input inspectable
0
Runtime trials
Not measured
3
Candidate actions
Fixture only
Action surface3 source candidates
SearchHuman UIInferred
CompareHuman UIInferred
AddHuman UIInferred
See the interaction patterns Explore the interaction model
Same task. Same app. Two paths.
The proof lab uses one synthetic catalog and one business-logic layer. Only the action surface changes.
Run the proof labBaseline
UI-only journey
Search → set filters → inspect → select → compare → add. Fully accessible and intentionally tool-free.
WebMCP
Structured journey
Search with constraints → compare stable IDs → update the same cart → verify visible state.
WebMCP Lift
Success, actions, elapsed time, retries, human intervention, and assertion coverage combine into a transparent −100 to +100 result.
Presence ≠ readiness
Four states worth separating.
Present
A source hint or registered object exists.
Ready
Contracts are narrow, distinct, state-aware, and safe.
Successful
A representative journey reaches its observable postconditions.
Safe
The same journey respects authorization, confirmation, and data boundaries.
Learn + monitor
The field guide now lives beside the lab.
Build from source-linked explainers and how-tos, then follow meaningful WebMCP changes without mixing them up with broader MCP ecosystem news.
WebMCP vs. MCP: what runs in the browser, what runs elsewhere
A practical map of where browser-exposed WebMCP tools fit within the much broader Model Context Protocol ecosystem.
Read 7-minute guideIs your web app ready for WebMCP? Run a baseline audit first
Start with the workflows, semantics, and verification signals your application already exposes before deciding which actions deserve tools.
Read 8-minute guideLatest signal
WebMCP explicitly brings headless handoffs into scope
Commit 50c4b7f moves headless browsing scenarios into the proposal's stated goals when page-exposed tools support task completion or transitions between human-in-the-loop and headless experiences. It still excludes purely server-side autonomous workflows and continues to position WebMCP as complementary to backend MCP; WebMCP remains an experimental Community Group draft, not a W3C Standard.
See source-linked updates