Loading ToolBench
MCP server with file storage capabilities for the Storacha Network - Decentralized storage for AI systems
The server defines 3 tools with basic schemas and descriptions present. Naming follows verb_noun patterns (identity, retrieve, upload). Descriptions exist but are brief (48-172 chars, baseline 194 avg). Parameter schemas are present with type definitions, but parameter descriptions are minimal or absent for several fields. Error handling and output schemas are not documented. The server supports multiple transports (stdio, SSE, REST) but STDIO is the primary/default mode.
Returns the DID key of the Storacha agent loaded from the private key storage config.
Retrieve a file from the Storacha Network. Supports formats: CID/filename, /ipfs/CID/filename, or ipfs://CID/filename.
Upload a file to the Storacha Network. The file must be provided as a base64 encoded string. The file name should include the extension (e.g., "document.pdf") to enable automatic MIME type detection.
identity tool has empty input schema {}, no parameters documented, but description is vague about what 'DID key' and 'Storacha agent' mean to an LLM unfamiliar with the domain
No output schemas documented for any tool, LLMs cannot reason about what fields will be returned (e.g., does retrieve return file_content, file_metadata, encoding, error details?). Makes chaining and error handling impossible.
retrieve tool's 'useMultiformatBase64' parameter lacks description explaining what multiformat base64 is, when to use it, or why an LLM should choose this option
upload tool's 'delegation' and 'gatewayUrl' parameters are optional but lack clarity on defaults, format, or typical values, LLMs will guess and likely pass invalid strings
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | D | 53 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 58 | - | v1 |
No error handling guidance documented, no recovery paths, retryable vs fatal error classification, or actionable error messages shown in code. Tools return raw errors without context.
retrieve tool description mentions format variants (CID/filename, /ipfs/CID/filename, ipfs://CID/filename) but does not constrain via regex or enum, LLMs will hallucinate invalid formats
identity tool description is only 48 chars, too short to explain DID key concept, when to call it, or what to do with the result. LLMs lack context for selection.