Browser extension that bridges TypingMind to MCP (Model Context Protocol) servers, synchronizing MCP tools as TypingMind plugins
This server defines a single dynamic tool 'mcp_*' that acts as a generic proxy to an external MCP bridge. The tool has a vague name, minimal description, and inadequate schema documentation. The implementation shows attempted functionality but violates fundamental tool design principles: the tool name provides no semantic meaning about what it does, the description is generic and doesn't explain when to use it, and the input schema is dangerously under-specified with an overloaded 'data' parameter that accepts both strings and objects. The server is not an MCP server itself, it's a TypingMind extension that fetches tools from an external MCP bridge and converts them into TypingMind plugins. This architectural choice means tool definitions are dynamic/inferred rather than explicitly registered, which per the rubric caps individual tool scores at 50 maximum.
Dynamically generated tools from MCP server endpoints. Each tool is created from MCP tool definitions fetched from the MCP bridge and executed via HTTP POST to the bridge endpoint.
Tool name 'mcp_*' is not verb-based and provides zero semantic meaning. LLMs cannot infer what the tool does from the name alone. Pattern requires names like 'get_', 'create_', 'search_', 'update_', etc.
Tool description is generic boilerplate ('Dynamic MCP tools fetched from MCP bridge server...') and does not explain WHAT the tool does, WHEN to use it, or any prerequisites. At 107 characters, it barely meets minimum length but lacks LLM-optimized guidance.
Input schema is severely under-specified: a single 'data' parameter that accepts both 'string' and 'object' types is dangerously ambiguous. The description states 'Can be a string (which maps to the first required parameter) or an object with named parameters' but this is implicit magic, no LLM can reliably infer when to use each mode without examples, and examples in descriptions are explicitly forbidden by the rubric.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 0 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 12 | - | v1 |
Parameter descriptions are vague and fail to specify constraints. 'Input data for the MCP tool' tells an LLM nothing about format, range, length limits, or required structure. A description should state format (JSON object keys matching tool parameters), length limits, and required vs optional fields.
No documented output schema. LLMs do not know what fields to expect in responses, making it impossible to plan downstream operations or extract relevant data. The tool description omits any explanation of return values.
Tool definitions are dynamically fetched from an external MCP bridge rather than explicitly registered in the server code. This means tool existence is inferred from runtime behavior, not statically visible.
No error handling strategy documented. If the MCP bridge is unavailable, returns invalid data, or the requested tool fails, the LLM receives no actionable recovery guidance. Errors should categorize as retryable, user-fixable, or fatal.
Type union on 'data' parameter (string | object) without discriminator or conditional constraints. This violates the 'constrained-input' pattern, LLMs cannot reliably choose between two fundamentally different input modes based on implicit rules. Should either split into two tools (call_mcp_tool_with_string, call_mcp_tool_with_object) or require explicit mode selector.