Loading ToolBench
Agent that automates browser actions to retrieve EDI and credentials using Playwright and MCP tools
Critical gaps across naming, descriptions, and schemas. 21 tools defined but most lack proper parameter descriptions and output schemas. Tool names are inconsistent (navigate vs browser_navigate, click vs browser_click) creating confusion. Many descriptions are generic or missing context. No error handling guidance. Security concerns: get_customer_credentials exposes sensitive data without clear safeguards. Schemas present but incomplete, parameters lack type constraints and validation rules. No pagination support despite tools returning potentially large results.
Click on an element in the browser
Execute JavaScript in the browser context
Fill form fields in the browser
Handle browser dialogs (alert, confirm, prompt)
Hover over an element in the browser
Navigate to a URL using local Playwright browser
Press a keyboard key in the browser
Duplicate tool sets (navigate/browser_navigate, click/browser_click, etc.) create LLM confusion. No clear distinction between the two sets. LLMs will waste reasoning cycles deciding which to call.
Sensitive data exposure: get_customer_credentials returns passwords and user IDs without documented access controls, rate limiting, or audit logging. No error handling for unauthorized access.
Parameter descriptions lack validation rules. 'selector' params accept CSS selectors but no guidance on format, escaping, or error handling for invalid selectors. 'script' param in browser_evaluate has no safety constraints or injection warnings.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | <=2025-11-25 | v2 |
Resize the browser viewport
Take a screenshot of the current browser page
Take a snapshot of the current page state
Click on an element in the browser
Fill a form field with text
Fill multiple fields in a table row
Fetch company configuration from GCS and return the website URL for the given customer_name
Returns today's date in YYYY/MM/DD format
Retrieves credentials (company_code, user_id, password) for a customer. Returns sensitive data - use carefully.
Hover over an element in the browser
Navigate to a URL in the browser
Set the checked state of a checkbox or radio button
Set the browser viewport size
Take a screenshot of the current page state
No output schemas documented. Tools like fill_table_row and browser_fill_form return success/failure but structure is undefined. LLMs cannot plan downstream calls without knowing response format.
Tool naming inconsistency: get_client_edi is cryptic (EDI undefined in description). get_customer_credentials lacks context on when to use vs get_client_edi. No dependency hints in descriptions.