Loading ToolBench
MCP (Model Context Protocol) adapter extension for Pi coding agent
Critical issues with tool definitions severely limit this server's quality. The three tools lack proper schema definitions, have inadequate descriptions, and fail basic naming conventions. Only the 'noop' tool has a minimal input schema. The 'mcp' and 'mcpScript' tools are defined in package.json and .ts files respectively, but lack explicit, machine-readable schemas. Descriptions, while present, are verbose and do not follow LLM-optimized patterns (should be 50-200 chars; these range 50-180+ chars but lack parameter documentation). No parameter descriptions are visible for any tool. This server appears to be an adapter/gateway rather than a traditional tool provider, but even gateway tools must expose clear, documented interfaces.
MCP gateway — URL installation, server status, tool search/describe, auth, and single MCP tool calls. When a user supplies an MCP endpoint URL, install it with the install action. When one request needs several MCP calls with logic between them, use mcpScript. Non-MCP Pi tools should be called directly, not through mcp.
Execute multiple MCP tool calls with custom logic between them
A no-operation test tool
Tool names 'mcp' and 'mcpScript' do not follow verb_noun convention. Should be action-oriented (e.g., 'call_mcp', 'execute_mcp_script').
No parameter descriptions visible for 'mcp' tool. The description mentions 'install action', 'tool search/describe', 'auth', 'single MCP tool calls' but does not document what input parameters exist, what types they accept, or constraints.
Tool 'mcp' description (180+ chars) is verbose and does not clearly state WHAT it does, WHEN to use it, or output structure. LLM-optimized descriptions should be 50-200 chars and direct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Tool 'mcpScript' description (57 chars) is minimal and lacks critical context: what is the syntax? How does it differ from 'mcp'? What does it return? No parameter documentation.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract data without knowing response structure.
No error handling guidance. Tools lack descriptions of failure modes, recovery paths, or actionable error messages.
Tool 'mcp' has WRITE risk but no confirmation/dry-run step mentioned. Agents installing servers or making API calls should have safeguards against mistakes.