Muxed is a STDIO-only MCP proxy/aggregator that exposes 7 tools for discovering and invoking tools across remote MCP servers. Tool definitions are present and partially documented, but exhibit significant quality gaps: (1) Most tool descriptions are bare operational statements lacking guidance on when/why to use them; (2) Critical tools like 'call' expose partial schemas, the 'call' tool accepts a 'name' parameter but does NOT document the dynamic arguments that follow, forcing LLMs to guess at invocation patterns; (3) Parameter descriptions are minimal or absent in several tools (e.g., 'depth' in 'grep' and 'tools' lacks explanation of its algorithm or token-budget behavior); (4) No output schemas are documented, leaving LLMs unable to plan downstream composition; (5) Error handling is present but generic, no recovery guidance (e.g., if a tool call fails, should the agent retry 'call', or re-run 'grep' to find alternatives?). The tool set itself follows single-responsibility design (one verb per tool), and tool names are reasonably clear (servers, tools, grep, info, call, resources, read), but the definitions lack the depth required for confident autonomous invocation.

• Document the complete input schema for 'call', including the dynamic tool-specific parameters. Clarify that LLMs must run 'info' first to fetch the target tool's input schema, then construct the call with those parameters. Add explicit guidance: 'To call a tool: (1) Run info(name=server/tool) to fetch its schema. (2) Construct a call(name=server/tool, <param1>=<value1>, ...) with the parameters listed in the schema.'
• Add documented output schemas to every tool. For 'tools', explicitly state 'Returns: Array<{server: string, tool: {name: string, description: string, inputSchema?: object}}>'. For 'call', state 'Returns: {content: Array<{type: "text", text: string}>, isError?: boolean}'.
• Expand parameter descriptions. For 'depth', add: 'Number of nesting levels to retain in schemas. Deeper nodes are replaced with {_collapsed: true, _hint: "..."}. If omitted, depth is auto-selected to fit a typical token budget (~4000 tokens). Use --depth 1 to show only top-level fields.'
• Add dependency hints to descriptions. 'tools': 'List all tools across servers. For full input schemas, use include=schema. To find a tool by pattern, use grep() instead. To get detailed info and schema for a specific tool, use info(server/tool).'
• Document error cases and recovery guidance for each tool. Example for 'call': 'If the tool is not found, run grep(pattern=<tool_name>) to verify it exists and identify the correct server. If the tool call fails (server offline, invalid input), inspect the error message and retry after fixing parameters or waiting for server recovery.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry