MCP server for iOS and Android device simulation and interaction, providing accessibility tree inspection, element search, gesture control, and live preview capabilities
SimView provides 14 well-named tools with consistent verb-noun patterns (list_, start_, observe_, search_, perform_, get_, inspect_, save_, end_). All tools have descriptions (avg ~80 chars, within baseline 34-392 range). Input schemas are present with type definitions and enums for codec/platform. However, output schemas are entirely undocumented, no response structure is visible in the source. Parameter descriptions are minimal (often single phrases like 'The session ID') and lack actionable constraints (ranges, formats, dependencies). Error handling is absent from visible code. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Composition is sound, tools chain logically (list_devices → start_session → observe_accessibility → perform_tap, etc.). Missing: output documentation, parameter validation guidance, error recovery patterns.
End the current SimView session and disconnect from the device
Retrieve the complete element tree for the current screen with optional pagination
Get live video preview stream from the connected device
Get accessibility information for elements at a specific screen point
List available iOS and Android devices with optional filtering by platform and availability status
Capture the current accessibility tree snapshot from the connected device
Perform a long press gesture at specified coordinates
Output schemas completely undocumented. No visible response structure for any tool, LLMs cannot plan downstream calls or extract required fields (e.g., what does start_session return? sessionId? token? metadata?). This violates pattern:tool and pattern:response-shaper.
Parameter descriptions are minimal and lack actionable constraints. E.g., 'maxWaitMs' has no range guidance (1-60000?), 'limit' says '1-25, default: 10' but no explanation of why 25 is the cap. Missing format/dependency hints per pattern:tool-description.
No error handling guidance visible. Tools like perform_tap, type_text, and press_key are WRITE operations that can fail (invalid coordinates, no focused field, device disconnected) but no recovery patterns documented. Violates pattern:recovery-guide and pattern:error-classification.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2025-06-18+ | v2 |
Perform a swipe gesture from one point to another
Perform a tap gesture at specified coordinates on the device screen
Press a hardware or software key on the device
Save annotated review images and metadata for the current session
Search for UI elements in the accessibility tree by text, type, or other criteria
Start a new SimView session on a specified device with optional codec preference
Type text into the currently focused text field
No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite clear risk classifications in metadata. LLMs cannot distinguish safe reads from destructive writes without explicit hints. Violates current MCP spec (2026-07-28) tool annotation requirement.
save_review_images parameter 'annotations' is an array with no schema. What is the structure of each annotation? Required fields? Enum values? LLMs cannot construct valid payloads without this. Violates pattern:constrained-input.