Loading ToolBench
MCP server for Standard Compute — list models, check endpoint health, and read plan pricing from any MCP client (Claude Code, Claude Desktop, opencode)
Four read-only tools with basic descriptions and minimal schemas. Three tools have empty input schemas ({}), making them trivial. chat_probe has one optional parameter with a description, but no output schemas are documented anywhere. Descriptions are adequate (100-200 chars) but lack actionable guidance on when to use each tool or what to do with results. No error recovery guidance. No tool annotations (readOnlyHint, etc.). Naming is clear and verb-forward (list_, endpoint_, plan_, chat_), but composition is weak, tools are discovery/diagnostic only, not actionable.
Send one short test prompt through the endpoint to verify end-to-end completion works with your key. Uses few tokens.
Check that the Standard Compute endpoint is reachable and measure round-trip latency. Works without an API key (an auth error still proves reachability).
List the model ids the Standard Compute endpoint serves for your API key (GET /v1/models). Requires STANDARDCOMPUTE_API_KEY.
Fetch Standard Compute's current subscription plans and pricing (live from standardcompute.com/llms.txt — always current, never cached here).
Three tools (list_models, endpoint_health, plan_pricing) have empty input schemas ({}). This violates pattern:constrained-input and pattern:tool-description.
No output schemas documented. LLMs cannot plan downstream calls or extract structured data. chat_probe returns text with embedded fields (model, reply) but no schema.
No error recovery guidance. Tools return raw HTTP status codes and error text (e.g., 'HTTP 401') without actionable next steps. Per pattern:recovery-guide, errors must tell the LLM what to do: 'API key missing, set STANDARDCOMPUTE_API_KEY env var' is actionable; 'HTTP 401' is not.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
No tool annotations. Tools are read-only but lack readOnlyHint annotation. Per current MCP spec (2026-07-28), tool annotations enable client-side safety checks and UI hints.
chat_probe parameter 'prompt' lacks context on format/length. Description says 'max 200 chars' but does not explain what happens if omitted (defaults to 'Reply with the single word: ok'). Per pattern:tool-description, parameter descriptions must state expected format and defaults.