MCP server for controlling the browser via Page Agent extension
Page Agent MCP has 3 tools with minimal schema documentation and weak descriptions. execute_task lacks parameter descriptions entirely (only 'task' with type string, no guidance on format or constraints). get_status and stop_task have no input parameters documented. Descriptions are present but generic (10-50 chars), below the 194-char baseline for production tools. No output schemas visible. No error handling guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tool composition is reasonable (separate concerns), but schema and description quality are far below production standards.
Execute a browser automation task described in natural language. The Page Agent extension will control the browser to complete the task. Blocks until the task is complete.
Check the current status of the Page Agent hub connection and agent.
Stop the currently running browser automation task.
execute_task parameter 'task' lacks description. LLM cannot infer expected format, length, or constraints. Should specify: natural language format, max length, example task types.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. get_status should document returned fields (connection_status, agent_status, etc.). execute_task should document result structure.
No tool annotations present. execute_task is destructive (WRITE risk) and should have destructiveHint=true. get_status is read-only and should have readOnlyHint=true. stop_task is reversible and should be annotated.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Descriptions are 10-50 characters, well below 194-char baseline. 'Execute a browser automation task...' (50 chars) lacks context on when to use vs alternatives, prerequisites, or expected duration. Should be 100-200 chars with actionable guidance.
No error handling guidance. If execute_task fails (browser crash, timeout, invalid task), LLM has no recovery path. Should document: retryable errors (timeout), user-fixable errors (invalid task format), fatal errors (browser unavailable).