Loading ToolBench
An MCP server that reconstructs your working context after interruptions by analyzing git state, recently modified files, editor state, and shell history
Rewind has well-intentioned tool design with clear, domain-specific purposes (context recovery, snapshots, activity tracking). All 5 tools have descriptions (avg ~180 chars, within baseline 194). However, input schemas are minimal: most tools accept only 1-2 string parameters with basic Zod descriptions. Output schemas are entirely undocumented, responses are formatted as markdown text with no structured field definitions. Error handling is absent; no guidance on retryability, user-fixable vs fatal errors, or recovery paths. Parameter descriptions lack constraints (ranges, enums, format specs). Tool names are clear and verb-forward, but composition could be tighter (where_was_i and diff_snapshots both compare states; recent_activity and where_was_i both analyze history).
Compares two snapshots (or the latest snapshot vs current state) to show what changed between them. Useful for understanding what happened while you were away, or for reviewing changes between any two points in time.
Shows the current status of the Rewind agent: tracked directories, snapshot counts, and storage info. Useful for understanding what the agent knows about and managing its state.
Shows a timeline of recent development activity in a project directory. Combines git commits, file modifications, and shell commands into a chronological view. Useful for understanding what happened in a project over a time period, or for reviewing your own recent work.
Takes a point-in-time snapshot of your current working context. Captures git state, recently modified files, editor state, and shell history. Use this before stepping away from your work (meetings, breaks, end of day) so that 'where_was_i' can compare your previous state with the current state when you return.
Reconstructs your working context after an interruption. Analyzes git state (branch, uncommitted changes, recent commits, stashes, merge state), recently modified files, editor state, and shell history to tell you: - What you were working on and why (with confidence levels) - What's in progress (uncommitted work, open branches, merge conflicts) - Suggested next steps prioritized by urgency - A timeline of recent activity Use this when you return from a meeting, break, or context switch and need to reload your mental state. Pass the project directory to analyze.
Output schemas completely undocumented. All tools return markdown text wrapped in {content: [{type: 'text', text: ...}]} with no structured field definitions. LLMs cannot extract typed data for downstream tool chaining or programmatic processing.
Input schemas lack constraints. 'directory' parameter accepts any string with no validation for absolute paths, existence checks, or error guidance. 'hours' in recent_activity has no min/max bounds (could accept negative or absurdly large values). No enums, patterns, or format declarations.
No error handling or recovery guidance. Tools silently fail if directory doesn't exist, git repo is corrupted, or shell history is unavailable. No error messages tell LLM what to do next (retry, ask user, try alternative tool).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Functional overlap between where_was_i and diff_snapshots (both compare states) and between where_was_i and recent_activity (both analyze history). Tool composition could be simplified; unclear when to call which tool.
get_context_status has empty input schema ({}). Description is vague ('Shows the current status'). No documentation of what fields are returned or how to use this for agent state management.