Loading ToolBench
An MCP server that manages and serves personal context information with privacy-aware filtering and update capabilities
Single tool server with moderate definition quality. The updateContext tool has a clear action verb in its name and a descriptive description (92 chars), meeting basic standards. However, the schema lacks critical constraint information, parameter descriptions are minimal (just 1-2 sentences), and there is no output schema documentation. The tool lacks error recovery guidance and doesn't follow input validation patterns. Resources are implemented but tool composition is weak, no batch variants, no pagination, and no natural-language identifier support. Overall, this is below median for production MCP servers.
Updates a field in the personal context with a reason for the change, tracks metadata and change history
Missing output schema documentation. The tool returns text confirmation but does not document the expected response structure, fields, or format for downstream tool chaining.
No input validation or constraint documentation. The 'path' parameter accepts freeform strings with dot notation but provides no regex pattern, length limit, or enumeration of valid paths. LLMs cannot self-correct invalid paths.
Error handling lacks recovery guidance. Error responses return a generic message without actionable next steps. E.g., 'Error updating context: ...' tells the agent nothing about whether to retry, validate input, or escalate.
No confirmation step for destructive operations. The updateContext tool modifies state and maintains change history, but provides no dry-run or confirmation mechanism. Agents may overwrite critical personal context data without verification.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 11 | - | v1 |
Parameter descriptions are extremely brief. 'Path to the field to update (e.g., \'preferences.learning_style\')' and similar descriptions are under 70 chars and lack usage context, dependencies, or format constraints. The rubric baseline for param annotations is 72 chars; this is at the threshold but lacks depth.
No permission gating or scope declaration. The updateContext tool modifies personal context with no documented permissions, audit trails, or role-based access control. Sensitive personal data should be gated.
Resources implement instruction-based filtering but lack comprehensive documentation. The 'instruction.privacy' mechanism is opaque, LLMs cannot predict what data will be filtered without detailed field-level documentation.