A programmable MCP proxy that reshapes tools with natural language transform prompts
mcpblox exposes 4 tools, but only 3 are user-facing (echo, add, lookup); the 4th (_mcp_metadata) is synthetic server introspection. All three functional tools have descriptions and input schemas with parameter types and descriptions, meeting baseline structural requirements. However, the descriptions are generic and under-optimized for LLM selection. Most critically, the test-server.ts tools are defined in a minimal integration-test context and lack production-grade error handling, output schema documentation, and contextual guidance. The echo, add, and lookup tools are intentionally simple for testing, not production use. The _mcp_metadata tool is self-referential and provides server introspection rather than domain functionality. Overall, the tools demonstrate competent schema hygiene but lack depth in description quality, output documentation, and error recovery patterns expected of A-grade production tools.
Returns metadata about the mcpblox server including upstream URL, hidden tools, pass-through tools, and synthetic tools
Adds two numbers
Echoes back the input message
Looks up an item by ID
Tool descriptions lack context for LLM selection. 'Echoes back the input message' (31 chars) is too terse. No explanation of when to use this tool vs alternatives, expected output format, or error conditions. Production baseline: 50-200 chars with WHAT, WHEN, and dependencies.
Output schemas not documented. Tools return text content, but no structured output schema is declared. LLMs cannot plan downstream operations without knowing what fields to extract. Pattern baseline: 100% of A+ tools document return types.
No error handling or recovery guidance. If 'lookup' receives an invalid ID, the tool does not indicate what went wrong or how the LLM should recover. Error responses should categorize as retryable, user-fixable, or fatal.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 38 | - | v1 |
Tool naming lacks verb-noun clarity. 'echo', 'add', and 'lookup' are valid verbs, but descriptions do not clarify context or alternatives. Production baseline: 90% of A+ tools start with action verbs and are 18 chars avg (p10=10, p90=27). These names are short but under-scoped for a real MCP proxy.
_mcp_metadata output schema incomplete. Returns JSON-serialized metadata (upstream_url, hidden_tools, pass_through_tools, synthetic_tools) but does not declare the structure of 'hidden_tools', 'pass_through_tools', or 'synthetic_tools' fields. LLMs cannot reason about nested structures without formal schema.
Parameter descriptions are minimal. 'message' (Message to echo), 'id' (Item ID), 'a' and 'b' (First number / Second number) lack constraints, examples, or edge-case guidance. Production baseline: 94% of A+ tools have param descriptions averaging 72 chars with format/range/constraint details.