Loading ToolBench
MCP Server for IA Agent with Tools - provides autonomous goal tracking, database access, Excel integration, IDE file operations, Gmail connectivity, and semantic search via Pinecone
Server has 19 tools with basic schemas and descriptions, but significant gaps prevent production use. Most tools lack parameter descriptions (e.g., run_db_query has no description for the 'sql' param). Descriptions are present but generic (avg ~80 chars, below 194-char baseline). No output schemas documented. Error handling is minimal, most tools return plain strings without recovery guidance. Security concerns: run_db_query accepts raw SQL with only regex blocking (insufficient); credentials not shown but Gmail/Pinecone keys likely in env vars (unverified). Tool composition is reasonable (separate read/write ops), but no pagination, no structured output, no tool annotations. Per-tool scores average 42 across the 19 tools.
Mark a registered goal as completed.
Return metadata about the configured Pinecone index.
Fetch the full content of a single email by ID.
Return all registered goals and their current status.
List all available tables in the database.
List the contents of a directory inside the workspace.
List all sheet names in the currently active Excel workbook.
run_db_query lacks parameter description for 'sql' input and has insufficient SQL injection protection (regex-only blocking). LLMs cannot infer what SQL syntax is valid, and regex bypass is trivial.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., does list_autonomous_goals return goal_id?). Responses are plain strings, not structured JSON.
No pagination support on list_* tools (list_autonomous_goals, list_db_tables, list_excel_sheets, read_emails). Large result sets will blow context window and degrade LLM reasoning.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | <=2025-11-25 | v2 |
Run a semantic search against the Pinecone index using plain text.
Fetch emails from Gmail matching a query.
Read the value of a cell from the currently active Excel workbook.
Read the contents of a file inside the workspace.
Register a long-term goal for the agent to pursue autonomously.
Execute a SQL query against the project database.
Analyze a task error and suggest a corrective strategy.
Send an email via Gmail.
Fetch an email and return a structured prompt for the agent to draft a reply.
Write a value to a cell in the currently active Excel workbook.
Insert or update records in the Pinecone index.
Write content to a file inside the workspace. Creates the file and any missing parent directories if needed.
Error handling is minimal. Most tools return plain strings (e.g., 'Goal not found: goal_123') without recovery guidance. LLMs cannot determine if errors are retryable or what to do next.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe read-only tools from destructive ones (e.g., update_excel_cell vs read_excel_cell).