Item logo image for WebMCP Explorer

WebMCP Explorer

Corsen AI
5.0(

1 rating

)
ExtensionDeveloper Tools33 users
Item media 5 (screenshot) for WebMCP Explorer
Item media 1 (screenshot) for WebMCP Explorer
Item media 2 (screenshot) for WebMCP Explorer
Item media 3 (screenshot) for WebMCP Explorer
Item media 4 (screenshot) for WebMCP Explorer
Item media 5 (screenshot) for WebMCP Explorer
Item media 1 (screenshot) for WebMCP Explorer
Item media 1 (screenshot) for WebMCP Explorer
Item media 2 (screenshot) for WebMCP Explorer
Item media 3 (screenshot) for WebMCP Explorer
Item media 4 (screenshot) for WebMCP Explorer
Item media 5 (screenshot) for WebMCP Explorer

Overview

Inspect, test, debug and review the security of the WebMCP tools a website exposes to AI agents.

WebMCP Explorer is a developer workbench for discovering, inspecting, testing and reviewing the tools a website exposes to AI agents through WebMCP. Core inspection, manual calls, recipes and passive security review need no AI account or API key. INSPECT THE PAGE'S TOOLS Read names, descriptions, input schemas, annotations and frame origins. Track registrations and removals, inspect contract-quality findings in Audit and compare saved snapshots. Explorer reports incomplete discovery and distinguishes native tools, captured registrations, JavaScript emulation and declarative form inventories. Open shadow roots are supported; closed roots remain unobservable. RUN TOOLS AND SAVE REUSABLE CHECKS Fill a generated form or enter JSON arguments, then inspect the result, timing, execution engine and dispatch status. Save argument sets and checks on selected JSON fields. Combine reviewed calls into manual recipes with named inputs and explicit use of earlier results. Pause, stop or cancel a run. When a call's effects are unknown, Explorer stops automatic continuation and does not replay that action. This protects Explorer's dispatches; it cannot guarantee what a website's own callback does internally. REVIEW SECURITY WITH EVIDENCE Thirteen rule families (WMCP-SEC-001 to 013) cover prompt-injection indicators, hidden or look-alike characters, consequential actions without hints, read-only claims contradicted by the callback's own source, declarative forms that submit hidden business fields or ask for credentials, untrusted content without untrustedContentHint, exposedTo lists and frames, text far beyond Chrome's budgets and provider keys shipped in page scripts. Findings show their evidence, severity, source and suggested remediation. A provisional 0–100/A–F grade makes the calculation and unknown coverage visible; it is not a security certification. Each fix shows the observed contract and the proposed one as a diff. Test this fix locally applies the proposed description, schema and hints on Explorer's side only: Agent Lab and evals see them, the page is never modified. When a page exposes no tool, its own forms become starter contracts, without auto-submit and without password, card or file fields. Passive review invokes no tool and sends no network request. CHECK SITES YOU CONTROL For a site you explicitly declare you control, choose tools and run bounded active probes through the normal execution/history path: wrong types, missing or unknown properties, an empty string where a minimum length is declared, a number outside its bounds and an inert marker that shows whether a tool echoes its input, at most eight calls per tool. Read-only annotations are advisory, not proof that a callback has no side effects. A website's tools can change data or contact its services. On request, Explorer also reads the page's inline scripts and same-origin script files once, in memory, and reports provider secret key formats by provider, type and length, never the value. Save a security baseline and optionally re-check already-open, permitted tabs passively; findings from newly added rules are listed apart from regressions. Scan Origin, only on request, checks discovery-file conventions and observes CSP, CORS and Permissions-Policy response headers. Missing conventions and header observations alone do not establish a vulnerability. CONNECT A LOCAL MCP CLIENT The optional zero-dependency Node.js companion connects a trusted local MCP client to Explorer through authenticated IPv4 loopback. It requires Node.js 22+ and local setup; no public npm installation is assumed. Pair the browser session and explicitly share individual permitted tabs. Corsen Guard shows the exact tool, origin and JSON arguments for approval and denies unanswered requests on timeout. Automatic read-only mode is optional and still checks annotations and security findings. Navigation, revoked access and reconnection invalidate shared bindings. Unknown effects pause the bridge. The paired client may retain or forward shared tool data according to its own settings. EXPORT TESTS AND REPORTS Export contracts, results and diagnostics as JSON or Markdown. Generate reviewed Playwright TypeScript tests from recorded calls and recipes, with contract/output checks and an explicit run ID. Export security findings as Markdown or SARIF 2.1.0 with full rule descriptions and security severities. Generated tests can invoke real page tools when you run them; generation itself invokes nothing. SARIF carries observed runtime locations. For GitHub code scanning of a public page, webmcp.corsen.ai/ci/ generates a ready workflow. Preview evidence and remove sensitive arguments or results before sharing. OPTIONAL: USE YOUR OWN AI PROVIDER Connect a supported cloud provider with your own API key or use a compatible local model server. No provider is selected automatically. Check the endpoint and list its models. Agent Lab displays safe Markdown, local prompt starters, saved conversations, prompt profiles and an expandable tool trace. Compare model choices with eval suites and try clearly marked local contract or result overrides. Provider usage may cost money; local models require a running server and suitable hardware. Before a run, Explorer names the destination and explains which messages, tool schemas, arguments and results will be shared. CONTROL ACCESS AND KEYS Explorer installs without site access and asks for one origin at a time. Review or revoke permissions in Settings. API keys use browser-session memory by default. An optional passphrase-protected AES-GCM vault retains encrypted keys on this device; unlock it for each browser session and lock or delete it whenever needed. Passphrases are not saved, and configuration exports exclude keys and the vault. Clear saved work in Settings. Minimal hashed invocation receipts remain separate from history so clearing history cannot authorize a duplicate action. Instrumentation and JavaScript emulation are separate, off-by-default choices for a permitted site and require a reload. Emulation does not replace native browser security or submit HTML forms. Optional Agent access is also off by default and shares only the requesting corsen.ai page's own tools and results. Explorer has no activity-collection backend, advertising or telemetry. GET STARTED Open Explorer from the Chrome toolbar, allow a site you want to inspect and open Tools. Help includes a guide, glossary and bundled demo. Native execution requires WebMCP support in both the browser and the page; the diagnostics report observed capabilities. An absent API does not prove a disabled flag. Eight interface languages: English, French, Spanish, German, Japanese, Simplified Chinese, Hebrew and Arabic, including RTL layouts. Published by Corsen AI. Source code and the companion guide: https://github.com/CorsenAI/webmcp-explorer Licensed under Apache 2.0. WebMCP Explorer is independent and is not affiliated with or endorsed by Google or an AI provider.

Details

  • Version
    2.6.1
  • Updated
    November 16, 2026
  • Size
    880KiB
  • Languages
    8 languages
  • Developer
    Corsen AI
    58 Rue de Monceau Paris 75008 FR
    Website
    Email
    hello@corsen.ai
    Phone
    +33 7 45 11 48 61
  • Trader
    This developer has identified itself as a trader per the definition from the European Union and committed to only offer products or services that comply with EU laws.
  • D-U-N-S
    286675059

Privacy

Manage extensions and learn how they're being used in your organization

WebMCP Explorer has disclosed the following information regarding the collection and usage of your data. More detailed information can be found in the developer's privacy policy.

WebMCP Explorer handles the following:

Authentication information
Personal communications
Web history
User activity
Website content

This developer declares that your data is

  • Not being sold to third parties, outside of the approved use cases
  • Not being used or transferred for purposes that are unrelated to the item's core functionality
  • Not being used or transferred to determine creditworthiness or for lending purposes

Support

For help with questions, suggestions, or problems, visit the developer's support site

Search
Clear search
Close search
Google apps
Main menu