Loading ToolBench

Reading the current corpus…

recto MCP server · A · ToolBench

recto

MCP server for a self-hosted AI journal backend. Provides tools to create, search, and reflect on journal entries with AI-powered enrichment, tagging, and semantic search capabilities.

<=2025-11-25STREAMABLE_HTTP7 tools

82/100

Definition Quality85

Protocol Readiness75

SupportabilityNot measured-

Spec posture

Inferred effective spec <=2025-11-25·current baseline 2026-07-28

Static source inference · medium confidence · detected: stateful session

Deprecated protocol patterns detected

Summary

The Recto MCP server defines 7 tools with generally good naming conventions (all start with action verbs: create_, get_, list_, search_, reflect, add_, get_). Tool descriptions are well-written and contextual, explaining WHEN to use each tool. Input schemas are comprehensive with proper JSON Schema structure (types, descriptions, enums where appropriate). However, output schemas are NOT documented in the visible source code, a critical gap for production-grade tools. Error handling guidance is absent from the code. The prompts feature is enabled but not shown in the source excerpt. Overall, this is a solid B-grade implementation with clear room for improvement on output documentation and error handling.

Tools (7)

add_tagswriteauthsource verified83/100

Add tags to an existing journal entry. Use this when the user wants to categorize or label an entry.

create_entrywriteauthsource verified87/100

Create a new journal entry. Use this when the user shares thoughts, experiences, reflections, or anything they want to record in their journal.

get_entryread onlyauthsource verified87/100

Get a specific journal entry by its ID. Use this to retrieve the full details of a known entry.

get_summaryread onlyauthsource verified85/100

Get a summary of journal entries for a specific time period. Use this when the user asks for a recap or overview of a day, week, or month.

list_entriesread onlyauthsource verified83/100

List journal entries with optional filters. Use this to browse recent entries or find entries by tag, date range, or mentioned people.

reflectread onlyauthsource verified80/100

Generate an AI-powered reflection on journal entries. Use this when the user asks for insights, patterns, or a summary of their journal over a period.

Issues (5)

HIGHtool-output-schema ↗

Output schemas not documented. Clients cannot predict what fields list_entries, search_entries, reflect, and get_summary return. This forces LLMs to guess about available fields and breaks tool chaining.

list_entriessearch_entriesreflectget_summaryget_entry

HIGHrecovery-guide ↗

Error handling guidance absent. No recovery hints or error classification. If create_entry fails due to invalid mood or people array, the LLM has no guidance on how to retry or correct input.

create_entryadd_tags

MEDIUMconstrained-input ↗

Date range parameters (from, to) accept strings but lack format specification. Descriptions say 'ISO 8601' but don't clarify if partial dates (YYYY-MM) are valid, timezone handling, or whether boundaries are inclusive/exclusive.

list_entriesreflectget_summary

Recommendations

Document the output schema for each tool. Example for list_entries: 'Returns {entries: [{id, title, content, mood, tags, people, created_at}, ...], total_count, has_more}'. This enables chaining and prevents guessing.
Add error classification to tool descriptions. E.g., 'If the entry ID is invalid, a 404 is returned, try list_entries to find valid IDs. If the mood is unrecognized, a 400 is returned with valid moods listed.'
Specify 'mode' enum semantics in search_entries description: 'hybrid: combines full-text and semantic matching; keyword: exact phrase matching; semantic: meaning-based search regardless of wording.'
Add bounds and defaults to numeric parameters: 'limit: max results to return (1 - 100, default 10). Requesting >100 returns exactly 100.'
Clarify date format: 'from/to: ISO 8601 date-time (e.g., 2024-12-25T00:00:00Z). Partial dates (YYYY-MM) are not accepted. Boundaries are inclusive.'
Add tool annotations (readOnlyHint, destructiveHint, idempotentHint) per current MCP spec. E.g., create_entry and add_tags should have destructiveHint=false (they are additive), and all read tools should have readOnlyHint=true.
Document the 'style' parameter in get_summary: 'brief: 1-2 sentence recap; detailed: 3-5 paragraph summary with moods and themes.'
Provide a companion error recovery tool or pattern. E.g., describe fallback behavior: 'If search_entries returns no results, list_entries can enumerate all entries for browsing.'

Spec posture evidence

Inferred effective spec: <=2025-11-25.

Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained

Score history

Overall score trend

↑ 82 points across a rubric change (v1 → v2)

82/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	A	82	<=2025-11-25	v2
2026-03-09	F	0	-	v1