This boilerplate server has 2 tools with basic schemas and minimal descriptions. The 'add' tool has a proper schema with typed parameters and descriptions, but 'getApiKey' is trivial (empty input schema, minimal description). Descriptions are 10 - 18 characters, falling below the 34-char p10 baseline. Parameters lack context about valid ranges, constraints, and relationships. No error handling guidance. The server includes a prompt ('add_numbers') but this evaluation focuses on tools. Overall, this is a starter template lacking production-grade quality, appropriate for a boilerplate but not suitable for agent deployment without significant enhancement.
Add two numbers
Get the API key
getApiKey tool exposes API credentials as tool output. Best practice is server-side secret injection via environment variables with no tool parameter or explicit credential return. The tool should either not exist (manage secrets outside MCP) or return a success/failure status without the actual key.
Tool descriptions are critically short: 'Add two numbers' (18 chars) and 'Get the API key' (15 chars). Baseline is p10=34 chars. Descriptions lack context about WHEN to use each tool, what it returns, and any constraints. LLMs cannot distinguish between tools with vague descriptions.
'add' tool lacks parameter constraints: no min/max bounds on 'a' and 'b', no guidance on what 'two numbers' means (integers? floats? range?). Unbounded parameters invite absurd values (Infinity, -1e308) that break math or cause unintended behavior.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 40 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 34 | - | v1 |
'add' tool output is returned as plain text (String(a + b)) inside a 'text' content block. There is no structured output schema documenting what fields the LLM should expect. Agents cannot reliably parse or chain this output.
'getApiKey' has an empty input schema ({}). While this is technically valid for a parameterless tool, the function is high-risk: it returns a secret that should never be exposed to an LLM or logged. The tool should be removed entirely or redesigned to authenticate the caller and return only a success flag.
No error handling. Neither tool documents how to recover if the operation fails. 'add' could error on non-numeric input; 'getApiKey' fails silently if the environment variable is unset (returns 'API_KEY environment variable not set'). This string should be a structured error response guiding the LLM on what to do next.
Tool names 'add' and 'getApiKey' are inconsistent in style. 'add' follows verb_noun convention; 'getApiKey' uses camelCase and abbreviates 'API'. This inconsistency increases LLM confusion when selecting between tools. Standardize to either snake_case (get_api_key) or consistent camelCase (addNumbers, getApiKey).