Loading ToolBench
Codex adapter for Honcho MCP with local transcript sync
Codex-honcho provides 7 tools with complete input schemas and descriptions. Naming follows verb_noun convention (search, create_conclusion, list_conclusions, delete_conclusion, get_context, get_representation, get_config). Descriptions are present and contextual (avg ~120 chars), explaining WHAT each tool does and WHEN to use it. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Parameters lack type constraints (e.g., 'scope' enum is present but 'limit' and 'page' lack min/max bounds). Error handling is minimal, no recovery guidance or actionable error messages visible in source. The server omits tool annotations (readOnlyHint, destructiveHint) despite having clear read/write/destructive semantics. Composition is sound: each tool has one responsibility, and search/list/get tools chain naturally. Missing: per-parameter descriptions for optional params (page, size, max_conclusions), output schema documentation, and error classification.
Save a key insight or biographical detail about the user to Honcho's memory
Delete a conclusion from Honcho's memory by ID. Use list_conclusions to find the ID first.
Get the current codex-honcho shim configuration (read-only).
Retrieve the full context object (representation + peer card) from Honcho for the current user. Scoped by observation mode.
Retrieve the user's representation string from Honcho. Lighter-weight than get_context.
List conclusions Honcho has saved about the user. Use this to review what is remembered before creating duplicates, or to find IDs for deletion.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for downstream tool chaining or data extraction.
Numeric parameters lack bounds. 'limit' and 'page' have no min/max constraints; 'max_conclusions' unbounded. LLMs may pass absurd values (limit=999999, page=-1).
Tool annotations missing. Despite clear semantics (search/list/get are read-only; delete_conclusion is destructive), no readOnlyHint or destructiveHint declared. Agents cannot infer safety properties.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Search across messages using semantic search. Defaults to the current session; use scope='workspace' to search across all sessions.
Error handling not visible. No recovery guidance, error classification, or actionable error messages in source. Agents cannot self-correct on failures.
Optional parameters lack descriptions. 'page', 'size', 'max_conclusions' have no guidance on defaults, ranges, or semantics. LLMs must infer from names alone.