Skip to content
Skills Library
Programming · Testing

verify

/verify

Check that a locally running site's WebMCP tools register and execute correctly. Use to verify WebMCP tools, confirm document.modelContext tools appear on the right pages and auth states, or test @nekuda/webmcp-sdk tools in a browser.

Repository
webmcp-kit ↗
Licence
MIT
Last repository change
23.10.2026

Assessment and security

98/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene13 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
1
Deprecation

No signals detected. Repository changed on 23.10.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Check that a locally running site's WebMCP tools register and execute correctly. Use to verify WebMCP tools, confirm document.modelContext tools appear on the right pages and auth states, or test @nekuda/webmcp-sdk tools in a browser.

Arguments

/verify [base URL of the running site]
Original author description

Check that a locally running site's WebMCP tools register and execute correctly. Use to verify WebMCP tools, confirm document.modelContext tools appear on the right pages and auth states, or test @nekuda/webmcp-sdk tools in a browser.

How it works

  1. Before you start
  2. Packaged browser driver
  3. Ladder
  4. Report — one state per tool
Original SKILL.md headings · 78 lines, 1,046 words
SKILL.md preview
---
name: verify
description: Check that a locally running site's WebMCP tools register and execute correctly. Use to verify WebMCP tools, confirm document.modelContext tools appear on the right pages and auth states, or test @nekuda/webmcp-sdk tools in a browser.
argument-hint: "[base URL of the running site]"
---

# WebMCP Kit — verify

Runtime check that a site's WebMCP tools register and work. Runs standalone, and is the ladder the `implement` skill uses in its verify step.

## Before you start
- The site must run locally. Start it, its services and its seeded test accounts the way the repo documents (README, `AGENTS.md` or contributing guide, package scripts); take machine facts, such as a browser this machine already runs, from its setup notes or the developer.
- Use the browser driver shipped in this plugin, or its npx form when this skill came from the AgentLane MCP (see *Without the plugin*). Do not ask for MCP configuration, a global `chrome-devtools` install, Playwright, or manual browser driving.

## Packaged browser driver

Resolve the plugin root once. Use `CLAUDE_PLUGIN_ROOT` or `PLUGIN_ROOT` when the host exports one. Otherwise, take the absolute base directory shown for this skill (the directory containing this `SKILL.md`) and go up two directories. Do not guess from the repository checkout.

Run one driver session for the whole ladder:

```sh
plugin_root="${CLAUDE_PLUGIN_ROOT:-${PLUGIN_ROOT:-}}"
if [ -z "$plugin_root" ]; then
  skill_dir="<absolute base directory for this skill>"
  plugin_root="$(cd "$skill_dir/../.." && pwd -P)"
fi
webmcp="${plugin_root}/scripts/webmcp.sh"

"$webmcp" browser status
"$webmcp" browser start
"$webmcp" browser status
"$webmcp" browser new_page "<local URL>" --output-format=json
"$webmcp" browser list_webmcp_tools <pageId> --output-format=json
"$webmcp" browser execute_webmcp_tool <pageId> <toolName> --input '<JSON object>' --output-format=json
"$webmcp" browser take_snapshot <pageId> --output-format=json
"$webmcp" browser list_console_messages <pageId> --output-format=json
"$webmcp" browser list_network_requests <pageId> --output-format=json
"$webmcp" browser stop
```

On Windows, apply the same fallback in PowerShell and use the packaged command entry:

```powershell
$pluginRoot = if ($env:CLAUDE_PLUGIN_ROOT) { $env:CLAUDE_PLUGIN_ROOT } elseif ($env:PLUGIN_ROOT) { $env:PLUGIN_ROOT } else { (Resolve-Path "<absolute base directory for this skill>\..\..").Path }
$webmcp = Join-Path $pluginRoot "scripts\webmcp.cmd"
& $webmcp browser status
& $webmcp browser start
```

Run the remaining ladder commands through `& $webmcp browser ...` and always end with `& $webmcp browser stop`.

`start` resolves the official `chrome-devtools-mcp@1.10.1` (pinned), launches one host-session-scoped isolated headless Chrome 150+ with both `--category-experimental-webmcp` and `--enable-features=WebMCP`, and disables usage statistics and CrUX while redacting network headers. Set `WEBMCP_BROWSER_SESSION_ID` to any unique value when concurrent runs share one workspace; otherwise the entry scopes by the Claude/Codex session when available, then the workspace. It first uses a supported installed or cached browser; if none exists, on a platform published by Chrome for Testing, it downloads stable Chrome into the user cache and reuses it later. You may pass `--executable-path <path>` for a known safe test browser. On an unsupported host such as Linux arm64, install a native Chrome/Chromium 150+ and set `WEBMCP_BROWSER_PATH` instead.

When the machine's setup notes name a dedicated test browser that already runs, is WebMCP-capable and exposes the Chrome DevTools Protocol (a remote, CI or headless machine, often with no Chrome to launch), set `WEBMCP_BROWSER_URL` to its endpoint, for example `http://127.0.0.1:9222`. `start` then attaches to it instead of launching Chrome, and every other command is the same. Point it only at that dedicated test browser, never the developer's own browser or profile: state-changing calls run in whatever session it holds.

If `start` fails, report its exact Node/npm, download, architecture, shared-library, or launch error and mark browser-dependent tools **could-not-verify**. Never use `sudo`, a system package manager, or a normal Chrome profile. After an interrupted run, use `status`, then `stop`, before retrying. Always stop the driver in cleanup, including after a failed rung.

### Without the plugin

When this skill came from the AgentLane MCP, there is no plugin root and no `scripts/webmcp.sh`. Run the same pinned driver through npx (Node 20.19+, 22.12+ or 23+). Choose one session id once, such as `webmcp-verify-<workspace folder name>`, and pass that same literal value to every command of the ladder: each command may run in a new shell. Run every command in full with a minimal environment, so the fetched package sees no tokens:

```sh
env -i PATH="$PATH" HOME="$HOME" npx --yes --package=chrome-devtools-mcp@1.10.1 -- chrome-devtools start --headless=true --isolated=true --category-experimental-webmcp=true --chrome-arg=--enable-features=WebMCP --performance-crux=false --usage-statistics=false --redact-network-headers=true --sessionId=<session id>
env -i PATH="$PATH" HOME="$HOME" npx --yes --package=chrome-devtools-mcp@1.10.1 -- chrome-devtools new_page "<local URL>" --output-format=json --sessionId=<session id>
env -i PATH="$PATH" HOME="$HOME" npx --yes --package=chrome-devtools-mcp@1.10.1 -- chrome-devtools list_webmcp_tools <pageId> --output-format=json --sessionId=<session id>
env -i PATH="$PATH" HOME="$HOME" npx --yes --package=chrome-devtools-mcp@1.10.1 -- chrome-devtools stop --sessionId=<session id>
```

The other ladder commands take the same form. Behind a proxy or a private certificate authority, also pass the machine's proxy and `NODE_EXTRA_CA_CERTS` variables through `env -i`. It needs an installed Chrome 150+; pass `--executable-path=<path>` to `start` when it is not the default Chrome. To attach to a dedicated test browser instead, `start` takes `--browserUrl=<endpoint> --category-experimental-webmcp=true --performance-crux=false --usage-statistics=false --redact-network-headers=true` and no other flags. For the Explorer, a cached browser download and the driver's other checks, install the WebMCP Kit plugin (`/plugin marketplace add nekuda-ai/webmcp-kit`, then `/plugin install webmcp-kit@nekuda`; Codex: `codex plugin marketplace add nekuda-ai/webmcp-kit`, then `codex plugin add webmcp-kit`).

## Ladder
1. **Boot.** Site starts; the baseline page renders; no *new* console errors versus a clean load.
2. **Discover.** On each declared page and auth state — check anonymous **and** signed-in where relevant — list the registered tools. Confirm each tool appears where it should, is **absent** where it should not (logged-out account tools, wrong-role tools), and unregisters on its declared exits (empty cart, logout). Navigation is the expensive unit: visit each page × auth state **once** and settle every tool's expectations for that page in that single pass — never one navigation per tool.
3. **Invoke read-only.** Call read-only tools with sample inputs; check the returned data **and** the visible ui_effect. Batch by page: invoke a page's tools in the visit that discovered them.
4. **Invoke state-changing.** Only against a positively confirmed local, development or seeded environment, with an explicit go-ahead; production and third-party services stay untouched. Confirm the persisted change by reading it back, never by a 2xx alone. A handoff tool is verified by the visitor landing on the handoff step with their choices kept; the card or credential step stays the visitor's. No safe way to invoke → report could-not-verify.

## Report — one state per tool
- **verified** — registered and invoked as declared (note the rung reached).
- **failed** — did not register or errored; must be fixed or dropped, never shipped.
- **could-not-verify** — plausible but unproven (no browser, or no safe way to invoke); ships flagged.

Print a per-tool table and restate any could-not-verify items.

Versions

RepositoryScoreLicence
This page · nekuda-ai/webmcp-kit98MIT
yonatangross/orchestkit100MIT
yonatangross/orchestkit100MIT
NolwennD/gabier100Apache-2.0
ericrisco/rsc-harness100MIT
sentimony/agentic-workflow100MIT
JoDi-27/JoDi-SDD100MIT
marcel-bich/marcel-bich-claude-marketplace100MIT
kdlbs/kandev100AGPL-3.0

Database snapshot: 13.11.2026 · package version b779a342b0a4 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 90implement
    Add WebMCP tools to a website codebase so browser agents can act through the site's own logic instead of scraping the page. Use when asked to make a site agent-ready or WebMCP-ready, add WebMCP or document.modelContext tools, or implement site tools with @nekuda/webmcp-sdk.
  • 100connect-existing-tools
    Connect a site's existing WebMCP tools to AgentLane without rebuilding them. Use when a codebase already registers document.modelContext tools — natively or through any SDK — and the developer wants them visible in AgentLane, migrated onto @nekuda/webmcp-sdk with durable stable…
  • 90webmcp-site-tools
    Design, implement, audit, and test WebMCP Site tools exposed by an existing web application through document.modelContext.registerTool. Use when an app should offer agent-facing actions in the live signed-in page or when Agent Graph needs a semantic browser verification lane. Do…
  • 100webmcp
    This skill should be used when the user asks to make a website or web app "WebMCP-compatible", "agent-ready", or usable by in-browser AI agents, or mentions WebMCP, document.modelContext, navigator.modelContext, registerTool, declarative form tools (toolname/tooldescription), or…
  • 100webmcp
    Make a website usable by in-browser AI agents with WebMCP. Plan which paths to expose, then ship them as declarative form annotations, dedicated imperative tools, or a bridge to an existing MCP server. Use when the user mentions WebMCP, document.modelContext,…
  • 100webmcpx
    Add WebMCP tools to a website and check them. Use when a user wants AI agents to use their site, asks about WebMCP, navigator.modelContext or document.modelContext, registerTool, or wants to lint/validate WebMCP tool definitions.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.