Momento MCP server implementation for cache and topic operations
The Momento MCP server has 6 tools with complete schemas and descriptions visible in the source code. Tool naming follows the verb_noun pattern (get, set, list-caches, create-cache, delete-cache, publish) which is excellent. Descriptions are present for all tools and parameters. However, there are significant issues: (1) output schemas are completely undocumented, the code returns generic text-based responses with no structured schema definition; (2) parameter descriptions lack actionable constraint details (e.g., 'value' in set/publish doesn't specify encoding, size limits, or format); (3) error handling is present but returns unstructured text responses instead of categorized, recovery-focused errors; (4) no response field documentation means LLMs cannot plan downstream tool chaining; (5) STDIO transport limits protocol readiness. The server is functional but falls short of production-grade tool design standards.
Creates a new cache
Deletes a cache and all contained values from your Momento account
get a key-value pair from the cache
Lists all cache names in your Momento account
publish a value to a topic
set a key-value pair in the cache
Output schemas completely undocumented. All tools return generic text responses via tool response content arrays with no structured schema or field definitions. LLMs cannot infer response fields for downstream tool chaining.
Parameter descriptions lack actionable constraints. 'value' param in set/publish does not specify: encoding (UTF-8?), max size, format restrictions, or whether null/empty is allowed. Similar gaps in 'key' and 'name' parameters.
Error handling returns unstructured text with status labels (SUCCESS, ERROR, MISS, UNKNOWN RESPONSE) instead of categorized, recovery-focused errors. No guidance on what to do next (retry vs ask user vs abandon).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 45 | - | v1 |
Tool descriptions are generic and lack context on when to use each. 'get a key-value pair from the cache' does not explain: when to use vs alternatives, what happens on miss/hit, or performance implications. No discovery guidance.
Pagination not implemented. list-caches returns all caches with no limit, offset, or next_cursor. A large account with hundreds of caches would return all in one response, wasting tokens and risking context window overflow.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Risk metadata is assigned in the spec but not declared in the MCP server, leaving LLMs without guidance on safe retry behavior.