Loading ToolBench
MCP server that provides screen recording summarization capabilities, extracting chapter markers and actionable steps from video recordings
This server exhibits severe quality deficiencies across nearly all dimensions. The MCP server implementation in mcp_server.py is non-functional, it merely prints a JSON status message and does not implement the MCP protocol at all. No actual tool registration, no request/response handling, no capability negotiation. The client.py file contains a standalone Python class unconnected to any MCP machinery. Tool definitions are inferred from example usage rather than explicitly registered. The two tools lack proper descriptions (under 20 characters for one, generic/vague for the other), have no input parameter validation or type constraints, and no documented output schemas. Error handling is absent. No error recovery guidance exists. The server cannot be invoked by any MCP client because it does not implement the MCP transport protocol.
Execute skill action for dynamic browser session recording summarization
Summarize a screen recording by extracting chapter markers, actionable steps, and generating an interactive player embed
MCP server implementation is non-functional. mcp_server.py contains only a print statement and does not implement the MCP protocol specification at all. No actual tool registration, no request/response message handling, no capability negotiation with clients.
Tool definitions are inferred from example_usage.py rather than explicitly registered in the MCP server. Tool 'execute_skill_action' is never actually implemented, it appears only in a printed JSON status message. Per HARD SCORING RULES, tools without explicit registration cap at score 50; this one has zero actual definition.
Tool 'execute_skill_action' has description under 20 characters ('Execute skill action for dynamic browser session recording summarization' is only slightly longer but is vague and generic). Per HARD SCORING RULES, description score capped at 0-20. No clarity on when to use this vs summarize_screen_recording.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 20 | 2026-07-28+ | v2 |
No input parameter schemas are formally defined for either tool. Tool 'execute_skill_action' has zero visible parameters. Tool 'summarize_screen_recording' has two parameters (video_stream_url, capture_mouse_heatmaps) but no JSON Schema type definitions, no constraints, no validation rules. Per HARD SCORING RULES, schema score MUST be 0 when no schema is visible.
Output schemas are undocumented. The example in client.py shows a mock return object with fields (summary_session_id, video_duration_seconds, chapter_markers, actionable_step_count, interactive_player_embed_url), but no formal JSON Schema is declared. LLMs cannot reliably parse unstructured output or plan downstream tool calls.
Parameter descriptions are missing or inadequate. 'capture_mouse_heatmaps' lacks any description explaining why an LLM should set it to true or false, what the implications are, or what 'heatmaps' means in this context.
No error handling, validation, or recovery guidance. The client.py method returns a hardcoded mock object with no error paths, no timeouts, no handling of malformed URLs, missing videos, or failed processing. A real agent would have no guidance on what to do if the video is inaccessible.
Tool naming ambiguity. 'execute_skill_action' is overly generic and vague, it does not convey what action is executed or when to use it instead of summarize_screen_recording. Per pattern:tool-naming guidance, names must be specific action verbs. 'execute_skill_action' violates the verb_noun convention and uses a non-descriptive generic verb.
No pagination or result limits documented. 'chapter_markers' in the output is returned as an array with no indication of limits, pagination support, or handling of videos with hundreds of markers. Per mxe:enforce-result-limits, even example-based results should cap output and offer pagination hints.
Client.py is a standalone class with no MCP message handling. The DynamicBrowserSessionRecordingSummarizerClient is a plain Python class that calls a mock method returning hardcoded data. It is not wired into the mcp_server.py at all, making it impossible for an MCP client to invoke these tools through the standard protocol.