Add webmcpevals binary for live WebMCP tool schema loading

AWESOME_WEBMCP.md

@@ -17,6 +17,8 @@ A curated list of awesome WebMCP demos.
   - **Example Prompt:** "I want to book a flight from New York to Los Angeles for two people on next Thursday."
 - [Animal Viewer](https://65s6dw.csb.app/) - A simple codesandbox demo page that shows either a dog or a cat image.
   - **Example Prompt:** "Show me a dog on this page"
+- **Moving Beyond Screen Scraping**: A hands-on example of using WebMCP to create an agentic first experience with 10x fewer tokens
+  - [Article](https://lnkd.in/daPRAtMX) | [Code](https://lnkd.in/dkZ3Jizn)
 
 ## Contributing
 

evals-cli/README.md

@@ -13,16 +13,19 @@ A TypeScript framework for evaluating the tool-calling capabilities of Large Lan
 The project is structured as follows:
 
 - `src/`: Source code.
-    - `bin/runevals.ts`: Main entry point that sets up the backend and runs the evaluation loop.
-    - `backend/`: Implementation of LLM backends (e.g., `googleai.ts`, `ollama.ts`).
-    - `types/`: TypeScript definitions for tools, messages, and evaluations.
+  - `bin/runevals.ts`: Entry point that loads tool schemas from a JSON file and runs the evaluation loop.
+  - `bin/webmcpevals.ts`: Entry point that loads tool schemas live from a browser page via the WebMCP API.
+  - `backend/`: Implementation of LLM backends (e.g., `googleai.ts`, `ollama.ts`).
+  - `browser/`: Browser automation for WebMCP tool discovery (`webmcp.ts`).
+  - `types/`: TypeScript definitions for tools, messages, and evaluations.
 - `examples/`: Detailed examples and test data.
     - `travel/`: A travel agent example containing `schema.json` and `evals.json`.
 
 ## Prerequisites
 
 - Node.js (v18+ recommended)
 - A Google AI Studio API Key (for Gemini models)
+- Chrome Canary 146+ with the `#enable-webmcp-testing` flag enabled (for `webmcpevals` only)
 
 ## Setup
 
@@ -51,46 +54,70 @@ The project is structured as follows:
 
 ## Usage
 
-### Running the Travel Example
+### `runevals` — file-based tool schemas
+
+Loads tool schemas from a local JSON file.
 
 ```bash
 node dist/bin/runevals.js --model=gemini-2.5-flash --tools=examples/travel/schema.json --evals=examples/travel/evals.json
 ```
 
-### Running evals with Ollama
+With Ollama:
 
 ```bash
 node dist/bin/runevals.js --model=qwen3:8b --backend=ollama --tools=examples/travel/schema.json --evals=examples/travel/evals.json
 ```
 
+| Argument    | Required | Default            | Description                           |
+| ----------- | -------- | ------------------ | ------------------------------------- |
+| `--tools`   | Yes      | —                  | Path to the tool schema JSON file     |
+| `--evals`   | Yes      | —                  | Path to the evals JSON file           |
+| `--backend` | No       | `gemini`           | Backend to use (`gemini` or `ollama`) |
+| `--model`   | No       | `gemini-2.5-flash` | Model name                            |
+
+### `webmcpevals` — live tool schemas via WebMCP
+
+Launches Chrome Canary, navigates to the given URL, and retrieves tool schemas live from the page via `navigator.modelContextTesting.listTools()`. Requires Chrome Canary 146+ with the `chrome://flags/#enable-webmcp-testing` flag enabled.
+
+```bash
+node dist/bin/webmcpevals.js --url=https://example.com/my-webmcp-app --evals=examples/travel/evals.json
+```
+
+| Argument    | Required | Default            | Description                           |
+| ----------- | -------- | ------------------ | ------------------------------------- |
+| `--url`     | Yes      | —                  | URL of the page exposing WebMCP tools |
+| `--evals`   | Yes      | —                  | Path to the evals JSON file           |
+| `--backend` | No       | `gemini`           | Backend to use (`gemini` or `ollama`) |
+| `--model`   | No       | `gemini-2.5-flash` | Model name                            |
+
 ## Argument Constraints
 
 You can use constraint operators to match argument values flexibly. A constraint object is identified when **all** its keys start with `$`.
 
 ### Supported Operators
 
-| Operator | Description | Example |
-|---|---|---|
-| **`$pattern`** | Regex match | `{"$pattern": "^2026-\\d{2}$"}` |
-| **`$contains`** | Substring match | `{"$contains": "York"}` |
-| **`$gt`**, **`$gte`** | Greater than (or equal) | `{"$gte": 1}` |
-| **`$lt`**, **`$lte`** | Less than (or equal) | `{"$lt": 100}` |
-| **`$type`** | Type check | `{"$type": "string"}` |
-| **`$any`** | Presence check | `{"$any": true}` |
+| Operator              | Description             | Example                         |
+| --------------------- | ----------------------- | ------------------------------- |
+| **`$pattern`**        | Regex match             | `{"$pattern": "^2026-\\d{2}$"}` |
+| **`$contains`**       | Substring match         | `{"$contains": "York"}`         |
+| **`$gt`**, **`$gte`** | Greater than (or equal) | `{"$gte": 1}`                   |
+| **`$lt`**, **`$lte`** | Less than (or equal)    | `{"$lt": 100}`                  |
+| **`$type`**           | Type check              | `{"$type": "string"}`           |
+| **`$any`**            | Presence check          | `{"$any": true}`                |
 
 ### Example
 
 ```json
 {
-    "expectedCall": {
-        "functionName": "searchFlights",
-        "arguments": {
-            "destination": "NYC",
-            "outboundDate": { "$pattern": "^2026-01-\\d{2}$" },
-            "passengers": { "$gte": 1 },
-            "preferences": { "$any": true }
-        }
+  "expectedCall": {
+    "functionName": "searchFlights",
+    "arguments": {
+      "destination": "NYC",
+      "outboundDate": { "$pattern": "^2026-01-\\d{2}$" },
+      "passengers": { "$gte": 1 },
+      "preferences": { "$any": true }
     }
+  }
 }
 ```