This is the primary set of tools for discovering and accessing new tools, methods, features, or capabilities. If a user asks the AI for a tool, method, feature, or capability that it does not currently have, it can use this tool to search for it in the MCP Finder registry. For example, if a user requests a browser tool, OCR, filesystem access, PDF parsing, or any other capability not present in the AI's current toolset, it can use this tool to search by relevant keywords. This allows the AI to expand its available MCP server toolset.
MCPfinder has 5 tools with clear verb-noun naming (search_, get_, add_, remove_, list_). All tools have descriptions (avg 95 chars, within baseline 34-392). Input schemas are present and use Zod validation with type definitions. However, parameter descriptions are sparse or missing in several tools, output schemas are not documented, and error handling lacks recovery guidance. The add_mcp_server_config tool has complex nested schemas but minimal guidance on when to use optional fields. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles (WRITE vs READ_ONLY).
Add or update an MCP server configuration in a client's configuration file
Get detailed information about a specific MCP server from the registry
List all MCP server configurations in a client's configuration file
Remove an MCP server configuration from a client's configuration file
Search for MCP servers in the MCPFinder registry by keywords or tags
Output schemas not documented. Tools return formatted text or JSON but LLMs cannot infer field structure for downstream chaining. E.g., search_mcp_servers returns 'text' blocks with 'ID: ...\nName: ...' but no structured schema.
add_mcp_server_config has optional nested object (mcp_definition) with conditional logic (command defaults if env/workingDir provided) but no guidance on when to omit vs include fields. LLMs will guess.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint). add_mcp_server_config and remove_mcp_server_config are marked WRITE but lack destructiveHint annotation to signal irreversibility to agents.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Error handling returns generic messages ('Failed to fetch from MCP Finder API') without recovery guidance. No actionable next steps for LLM when API is down or server not found.
Parameter descriptions in Zod schemas are minimal. E.g., 'query' in search_mcp_servers says 'Keywords to search for' but does not explain: search scope (name only? description? tags?), case sensitivity, OR vs AND logic, or max length.