Model Context Protocol server for ActivityWatch time tracking data
Mixed quality across 10 tools. Naming follows verb_noun convention consistently (e.g., list_buckets, run_query, add_category). Descriptions exist for all tools but vary in depth, some are clear and action-oriented (e.g., activitywatch_run_query: 'Run a query in ActivityWatch's query language'), while others are minimal (e.g., activitywatch_query_examples: 'Get examples of properly formatted queries'). Input schemas are present with proper JSON Schema structure, but documentation of output schemas is absent, no tool documents what fields are returned. Error handling exists in the main index.ts (makeSafeToolResponse wrapper), but guidance is generic ('Error: <message>') rather than recovery-oriented. No tool uses tool annotations (readOnlyHint/destructiveHint), though several tools (3 WRITE, 1 DESTRUCTIVE) are marked for these in metadata. No security-critical tools (add_category, update_category, delete_category) validate permissions or log audit trails. Parameters are generally well-typed with descriptions, though several lack examples or constraints (e.g., activitywatch_run_query's 'query' parameter uses a confusing array-of-strings schema with minItems=1, maxItems=1, suggesting it expects exactly one string despite being defined as an array).
Add a new category to ActivityWatch. The category will be appended to the existing list.
Delete an ActivityWatch category by its name path.
Get all ActivityWatch categories (classification rules). Returns the list of categories with their regex rules used to classify window events.
Get raw events from an ActivityWatch bucket
Get ActivityWatch settings. Can retrieve all settings or a specific key if provided.
Fetch window events that have NOT been matched by any category rule, grouped and sorted by total duration. Use this to discover what activities need new category rules. Returns the top N app+title combinations by time spent.
No output schemas documented. Tools declare input parameters but provide no formal documentation of response structures. LLMs cannot predict field names, types, or array structures they will receive, forcing manual parsing and increasing hallucination risk.
Tool annotations missing. Three write tools (add_category, update_category, delete_category) and one destructive tool (delete_category) lack destructiveHint/idempotentHint annotations. Without these, LLMs cannot determine which tools are safe to retry and which carry irreversible consequences.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | C | 60 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 45 | - | v1 |
List all ActivityWatch buckets with optional type filtering
Get examples of properly formatted queries for the ActivityWatch MCP server
Run a query in ActivityWatch's query language
Update an existing ActivityWatch category by its name path. Replaces the matching category with new values.
Confusing array schema for activitywatch_run_query 'query' parameter. Defined as array with minItems=1, maxItems=1, but description says 'MUST BE A SINGLE STRING' and example shows ["events = query_bucket(...)"] with one string. This mismatch will confuse LLMs about whether to pass a single string or an array.
Minimal error guidance. Error responses in makeSafeToolResponse return generic error messages ('Error: <error message>') without recovery hints. Per pattern:recovery-guide, errors must tell the LLM what to do next (e.g., 'User not found. Try search_users() with a partial name.').
No permission gates on destructive operations. delete_category and update_category tools perform irreversible changes without verifying user/agent authorization. Per pattern:permission-gate, destructive tools must verify calling user has authority.
No audit logging. Tools that modify state (add_category, update_category, delete_category) do not log who called them, what parameters were used, or what happened. Per pattern:audit-trail, agent-initiated actions must be traceable.
activitywatch_query_examples description is under 20 characters (59 chars). While not critically short, descriptions for discovery tools should explain what structure they reveal and when to call them (per pattern:tool-description).