Loading ToolBench
A policy-enforcing Streamable HTTP MCP bridge from ChatGPT to local Codex.
Critical gaps across all dimensions. Tool definitions lack proper schemas, descriptions are vague or missing, and parameter documentation is severely incomplete. Of 20 tools examined, most have minimal descriptions (10-50 chars, well below the 194-char baseline). Input schemas are either absent or severely underspecified, many tools show only partial parameter definitions with no type information. No evidence of output schema documentation. Error handling guidance is absent. Tool names lack clear verb-noun structure (e.g., 'codex_answer', 'codex_task', 'bridge_skill' are ambiguous about what they do). This server would not pass code review for production use.
App-only hydration operations
App-only mutation operations
Execute bridge skill operations
Manage bridge skills
Cancel activity operations
Update activity state and metadata
Manage Codex agents and their context
Answer a current ordinary Codex question in this conversation. Refresh codex_status kind=input after any intervening user deliberation or decision card and use only the exact questionRef that remains current. This cannot grant approvals, supply authentication secrets, or start another turn.
Missing or severely incomplete input schemas. Most tools show only partial parameter definitions (e.g., codex_task has 'selection' as bare object with no properties defined). No type information for most parameters.
Descriptions are too short (10-50 chars) and lack actionable context. Most tools have single-sentence descriptions that do not explain WHEN to use them, WHAT they return, or any prerequisites. Baseline is 194 chars; these are 80% below baseline.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2025-06-18+ | v2 |
Cancel running Codex jobs
Display dashboard of retained work and activity history
Create or retrieve decision cards for user interaction
Retrieve decision card results
Access diagnostic information for troubleshooting
Retrieve available model information and capabilities
Manage user settings and project configuration
Query status of Codex jobs, projects, and input
Steer execution of Codex operations
Execute a Codex task with model selection and project context
Retrieve task error information
Query task execution state
Tool names lack clear verb-noun structure. Names like 'codex_answer', 'codex_task', 'codex_steer', 'bridge_skill' are ambiguous about the action. LLMs cannot infer intent from the name alone. Should be 'answer_codex_question', 'execute_codex_task', 'steer_codex_execution', 'execute_bridge_skill'.
No output schema documentation visible. LLMs cannot plan downstream tool calls or extract required fields without knowing what each tool returns. All 20 tools lack documented return types.
Parameter descriptions are missing or trivial. Parameters like 'query', 'selection', 'operation', 'html' lack descriptions explaining what values are valid, what format is expected, or what the parameter controls.