Document hosting and encrypted agent memory with multi-tenant persistence, payment-gated via x402 protocol on Base network
DocDeploy MCP server has 13 tools with consistent naming (verb_noun pattern) and descriptions present for all tools. However, critical gaps exist: (1) Parameter descriptions are minimal or missing for many inputs, e.g., 'limit' and 'offset' in list_documents lack range constraints; (2) Output schemas are not documented anywhere in the provided source, LLMs cannot plan downstream calls without knowing what fields are returned; (3) The '_payment' parameter appears in all tools but is poorly described ('Optional x402 payment envelope') and represents a non-standard pattern that should be abstracted server-side, not exposed as a tool parameter; (4) Error handling guidance is absent, no recovery hints or categorization; (5) Tool descriptions average ~100 chars, below the 194-char baseline for A+ tools, lacking context on when to use each tool vs. similar ones (e.g., search_documents vs. list_documents distinction is unclear). Memory tools (remember, recall, forget) have domain-specific parameters (principal, domain, kind) that lack enum constraints or validation rules. Positive: naming is clear and action-oriented; all tools have descriptions; risk classifications are present.
Delete a specific memory by ID, permanently removing it from encrypted storage
Retrieve a full document by ID with all sections and metadata
Retrieve the document manifest containing metadata about all available documents
Retrieve memory cadence preferences (mode, triggers, per-domain settings) for a principal
Get the original presigned URL for downloading a document in its original format
Retrieve a specific section from a document by section ID
Output schemas not documented. LLMs cannot infer what fields are returned by any tool, preventing downstream tool chaining and forcing wasteful discovery calls.
'_payment' parameter exposed as tool input. Credentials and payment envelopes must never appear as parameters, they leak into logs and traces. Move to server-side secret injection via headers or environment.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
List memory metadata (without decrypted bodies) for a principal and domain with optional filtering
List all documents available to the authenticated tenant with pagination
Retrieve and decrypt memories matching principal, domain, and optional kind filter
Store encrypted memory for an agent, scoped to a principal and domain with optional metadata
Search documents by keyword, returning matching sections with context and presigned download URLs
Update memory cadence preferences (mode, triggers, per-domain settings) for a principal
Identity echo tool that returns the authenticated tenant, default principal, and memory cadence preferences
Parameter descriptions lack constraints. 'limit' and 'offset' in list_documents have no min/max; 'kind' in recall has no enum or validation rules. LLMs will pass invalid values.
Tool 'list' has generic name. Conflicts with common list_* pattern. Should be 'list_memories' to clarify it lists memory metadata, not documents.
No error handling guidance. Tools lack recovery hints (e.g., 'If document not found, try search_documents'). LLMs cannot self-correct on failures.