Free temporary object storage for AI agents. 2 hour expiry. Cache contexts, bridge multi-agent workflows, share data between agents and sessions.
This server presents two tools with moderate documentation but significant structural and design issues. Both tools have descriptions and input schemas present, but the schemas lack proper type definitions for critical fields, parameter descriptions are incomplete or missing, and the tools exhibit composition problems. The 'data' parameter in create_object has no type constraint, making it ambiguous what LLMs should pass. Neither tool includes output schema documentation. The naming is reasonable (verb_noun pattern) but the tool pair lacks proper separation of concerns, they form a simple read/write pair with no error recovery guidance. Error handling is minimal: the HTTP wrapper returns generic JSON-RPC errors without actionable recovery hints. The server is STDIO-only, which is a hard transport constraint, but even accounting for that, the definition quality is below baseline.
Save objects to ephemeral cloud storage - Store your reasoning state, large JSON structures, or any data to get a shareable URL. Perfect for: caching contexts before token limits, bridging multi-agent workflows, storing intermediate results, sharing data between sessions, or temporary data storage. Returns a shareable URL valid for 2 hours with auto-expiring data.
Retrieve data from cloud storage using its ID. Use this to read data that another agent (or yourself) saved previously.
Critical: 'data' parameter in create_object has no type definition. Schema declares {"description": "The data to save - can be an object, array, or string"} but omits 'type' field. This violates JSON Schema and leaves LLMs unable to validate input.
High: Output schemas are completely undocumented. create_object and get_object return JSON responses with fields like 'success', 'object_id', 'url', 'expires_in' for create, and 'success', 'data' for get, but these are never formally declared in the tool definition. LLMs cannot plan downstream calls or validate responses.
High: Missing parameter descriptions. 'agent_id' in create_object has a description, but 'intent' lacks detail on its importance for analytics/debugging. 'ttl_seconds' includes a range but no guidance on typical use cases. 'object_id' in get_object has only a bare identifier description, no examples or format guidance.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 48 | - | v1 |
High: Error handling is non-actionable. The HTTP wrapper's catch block returns a generic JSON-RPC error with a stack trace. There is no guidance on retryability, user-fixable vs. fatal conditions, or recovery steps.
Medium: Tool descriptions lack guidance on when to use each tool. 'get_object' says 'Retrieve data from cloud storage using its ID' but does not explain the relationship to create_object or what typical workflows look like. 'create_object' does list use cases (caching, bridging workflows, sharing data) but get_object mirrors none of that context. When should the LLM call it instead of a similar tool?'
Medium: The 'data' parameter is too permissive. It accepts 'object, array, or string' with no constraints on size, nesting depth, or content type. LLMs might pass 100MB JSON structures, infinite nesting, or non-serializable types. When a parameter has a regex pattern, length limit, or character restriction, state it in the description.'
Medium: No pagination or result limits documented. Although get_object retrieves a single object by ID, create_object returns a single response. If this service scales to multi-object returns in future, there is no guidance on pagination.