MCP server for posting daily reports to esa.io
Single tool 'times-esa' with complete JSON Schema and basic descriptions. However, naming violates verb-first patterns, descriptions are minimal (47 chars for tool, 57 chars for main param), and error handling provides no recovery guidance. The tool is explicitly defined and registered, but lacks LLM-optimization guidance typical of production agent tools. Parameters have types and descriptions (meeting baseline), but are sparse and generic. No output schema documentation. No tool annotations despite WRITE risk classification.
times-esaに日報を投稿します
Tool name does not start with action verb. 'times-esa' is a noun/service name, not a verb-first pattern. Should be 'submit_daily_report' or 'post_to_times_esa' to signal intent and enable LLM action inference.
Tool description is 47 characters, below recommended 50 - 200 character range for LLM-optimized descriptions. Current text 'times-esaに日報を投稿します' (posts daily report to times-esa) lacks context on WHEN to use the tool or what the user gets back. Should state: 'Submits a daily report entry to the team's times-esa workspace. Creates a new entry for today's date or appends to an existing entry if one already exists.'
No output schema documentation. Tool returns a TimesEsaPostResponse with Success (boolean), Message (string), and Post (EsaPost object). LLMs cannot infer downstream field names (e.g., whether to expect 'post_id', 'post_number', or 'post.number'). This breaks tool chaining and forces agents to guess response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 30 | - | v1 |
Error messages lack recovery guidance. Examples: 'text parameter cannot be empty' (no suggestion), 'ユーザーが投稿内容を確認したかどうか' (confirmation failure, but doesn't guide LLM to retry with confirmed_by_user=true). Per pattern:recovery-guide, errors should state 'Set confirmed_by_user=true to authorize posting before retrying.'
Destructive tool (WRITE risk) lacks confirmation workflow. No dry-run or explicit confirmation step. The 'confirmed_by_user' parameter is a workaround but not documented in the tool schema, it appears in code but not as an annotation. Per pattern:confirmation-request, tools that modify external state should support a two-step confirmation.
Parameter 'confirmed_by_user' is not idiomatic for natural LLM interaction. A user would say 'Post this report', not 'Post this report with confirmed_by_user=true'. Per pattern:chat-data-model, tool interface should match user intent. Consider a dedicated tool step or elicit confirmation via the LLM context, rather than exposing it as a parameter.
Environment variables (ESA_TEAM_NAME, ESA_ACCESS_TOKEN) are required but not validated at tool instantiation. Error 'ESA_TEAM_NAME または ESA_ACCESS_TOKEN が設定されていません' will surface only at runtime when the tool is called. Consider validating at server startup and returning a clear initialization error.
No rate limiting or debounce configuration documented in the tool schema. Code implements debounce logic (isDebounced, debounceConfig.Duration) but LLM has no visibility into rate limits or retry delays. Should document: 'Duplicate submissions within 5 seconds are rejected. LLM should wait 5+ seconds before retrying.'