Dynamically routes MCP requests to different MCP servers based on current directory path patterns. Maps filesystem paths to specific MCP server configurations and proxies requests transparently.
This MCP server acts as a proxy/dispatcher that forwards tool and prompt requests to backend MCP servers. All four tools have minimal definitions with significant quality gaps: descriptions are generic boilerplate stating only 'Forward [operation] request to backend server' without explaining what the dispatcher does, when to use it, or how to work with backend servers. Input schemas are present but severely incomplete, call_tool and get_prompt have sparse parameter descriptions lacking context about argument structure, mapping behavior, or error recovery. No output schemas are documented. The server provides no error handling guidance, authentication/security documentation, or composition patterns for selecting between backend servers. Tool names are adequate (list_*, get_*, call_*) but descriptions fail to guide LLM reasoning about dispatcher vs backend tool semantics.
Forward call_tool request to backend server
Forward get_prompt request to backend server
Forward list_prompts request to backend server
Forward list_tools request to backend server
All four tools lack meaningful descriptions. 'Forward [operation] request to backend server' (15-20 chars) is below the 10-character minimum but provides zero actionable context. Descriptions do not explain what the dispatcher does, when to invoke it vs calling backend tools directly, or how backend server selection works.
list_tools and list_prompts have NO input schema defined.
call_tool accepts 'arguments' as type 'object' with description 'Tool arguments', extremely vague. No guidance on what keys/values are valid, how to structure nested objects, or what happens if an invalid argument is passed. LLMs cannot infer proper argument structure.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 32 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 16 | - | v1 |
No output schemas documented for any tool. LLMs cannot know what fields to expect from list_tools, call_tool, list_prompts, or get_prompt responses, forcing them to make assumptions about downstream tool chaining.
No error handling guidance. What happens if a backend server is unreachable? If an invalid tool name is passed to call_tool? If the backend returns an error? No recovery instructions for the LLM.
Tool 'call_tool' accepts a 'name' parameter (tool name) but lacks guidance on discovery: how does the LLM know what tool names are available? Must call list_tools first? What if the name does not exist in the backend?
Tool 'get_prompt' accepts 'arguments' (type object, description 'Prompt arguments') but no clarity on structure, required keys, or how arguments map to backend prompt parameters. Undocumented parameter dependencies.
No composition guidance. What is the typical workflow? list_tools → call_tool? list_prompts → get_prompt? Does the dispatcher maintain state across calls or is each request independent? Unclear to LLMs how to chain these tools.