A meta-MCP server for discovering and using other MCP servers. Provides search, tool listing, and function execution capabilities for MCP servers.
InfiniteMCP has critical definition quality gaps. Three tools are present with input schemas, but descriptions are vague and lack actionable guidance for LLMs. The `search_mcp` tool has a reasonable description (70 chars), but `list_tools` and `execute_function` lack clarity on prerequisites, error conditions, and the structure of the 'config' object they depend on. Parameter descriptions are sparse, the 'config' parameter in both `list_tools` and `execute_function` references nested properties (one_liner, requires) but provides no guidance on what these fields actually mean or how to obtain them. The `parameters` field in `execute_function` has a generic description ('Function parameters as a JSON object') that provides no constraints, validation rules, or format guidance. No error handling guidance is present, LLMs will not know what to do if a config is invalid, a function execution fails, or a required environment variable is missing. The server exposes a read-only search/list workflow followed by execution, but lacks idempotency hints, retry guidance, and documentation of the side effects of `execute_function`. Output schemas are completely undocumented, LLMs cannot know what `search_mcp` returns, what fields `list_tools` returns, or what the response structure of `execute_function` will be.
Execute a specific function from an MCP server. Handles credentials and parameters.
Get all available functions/tools from a specific MCP server. Use the config from search_mcp results.
Search for MCP servers by functionality. Returns structured configs with commands and required credentials.
Missing output schemas for all three tools. LLMs cannot infer the structure of responses from search_mcp, list_tools, or execute_function. This forces LLMs to guess at field names and types, increasing hallucination risk.
Parameter 'config' in list_tools and execute_function is inadequately documented. The description references nested properties 'one_liner' and 'requires' but does not explain: (a) what one_liner contains (is it a string, array, command template?), (b) what requires lists (environment variable names?), (c) where to obtain this config object (only from search_mcp?), or (d) what happens if the config is malformed.
No error handling guidance. If execute_function fails (e.g., missing env vars, invalid function name, execution timeout), the description provides no hint for recovery. LLMs will not know whether to retry, ask the user, or call a different tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 39 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 39 | - | v1 |
No validation constraints on 'parameters' input to execute_function. The field accepts 'any JSON object' with no type, format, or allowed key constraints. This invites hallucinated parameter names and invalid values.
The 'env_vars' parameter in execute_function is marked optional (default: {}) but the description says keys should match the 'requires' field from config. If an env var is required but omitted, execution will fail. No guidance on which vars are mandatory vs optional.
search_mcp description states it 'Returns structured configs with commands and required credentials' but does not document the actual response structure. What fields are in the config? Are credentials redacted or exposed? Can the response be paginated?
No idempotency or side-effect documentation. execute_function is marked as WRITE risk but does not indicate whether calling it twice with the same params will execute the remote function twice (non-idempotent) or once (idempotent). Agents need this to know if retry is safe.