Self-hosted gateway that aggregates local (stdio) and remote (HTTP) MCP servers behind one OAuth-secured endpoint.
MCP Switch exposes a single proxy tool 'remote_tool_proxy' that dynamically forwards calls to upstream MCP servers. The tool definition is visible in apps/mcp-gateway/src/tools.ts but lacks critical quality markers. The description is generic and does not explain WHEN to use this tool or what it returns. Input schema is declared as 'Dynamic schema derived from upstream tool's inputSchema', this is not a concrete JSON Schema; it's a runtime coercion strategy. No output schema is documented. The tool accepts arbitrary upstream schemas and coerces types (number/boolean/array/object from strings), but this coercion logic is implementation detail, not part of the tool contract. Error handling is absent from the visible code. The tool name 'remote_tool_proxy' is generic and does not follow verb_noun convention (should be something like 'call_remote_tool' or 'invoke_upstream_tool'). This is a gateway/aggregator pattern, not a domain-specific tool, so it scores lower than a typical tool, it's infrastructure, not user-facing functionality.
Proxy tool that forwards calls to upstream MCP servers. Each remote tool is registered with a unique skillId and forwards arguments to the upstream server via the registry client.
Tool name 'remote_tool_proxy' is generic and does not follow verb_noun convention. Should be 'call_remote_tool', 'invoke_upstream_tool', or similar to clarify intent.
Input schema is not a concrete JSON Schema, it is described as 'Dynamic schema derived from upstream tool's inputSchema, coerced for type compatibility'. No actual schema object is visible. This violates the requirement that every tool must have a documented input schema.
Tool description is minimal and does not explain WHEN to use this tool, what it returns, or how it differs from calling upstream servers directly. Descriptions must be 10 - 1024 characters and answer: What does it do? When should the LLM call it? What does it return?
No output schema is documented. LLMs cannot plan downstream tool calls or extract data without knowing what fields to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
Error handling is not visible in the tool definition. No guidance on retryability, user-fixable errors, or recovery steps. Errors must tell the LLM what to do next.