Loading ToolBench
MCP server for trading stocks via Zerodha Kite Connect API. Provides tools to buy/sell stocks, check positions and holdings.
This server has critical definition quality gaps across most tools. Only 2 of 5 tools have descriptions above 20 characters. Parameters lack type constraints (no enums for trading symbols, no ranges for quantities). Output schemas are completely undocumented, tools return JSON.stringify() with no indication of field structure. The 'add' tool is vestigial and clutters the interface. Error handling is absent, all try-catch blocks log to console but return nothing on failure, leaving LLMs without recovery guidance. No tool has documented output structure. Security is compromised: access tokens are hardcoded in index.ts and the trading symbol parameter accepts any string without validation.
Addition tool for adding two numbers
Buy stock with specified trading symbol and quantity
Get current stock holdings
Get current open positions
Sell stock with specified trading symbol and quantity
No output schemas documented. Tools return JSON.stringify() with no indication of response structure, field types, or required fields. LLMs cannot plan downstream tool calls or extract specific data.
Parameters accept free-form strings with no validation. 'tradingsymbol' has no enum, regex pattern, or format constraint. This invites LLMs to hallucinate invalid stock symbols and cause failed orders.
Hardcoded secrets in source code. apiKey and access_token are visible in index.ts. If this repo is public, credentials are compromised. Secrets must never appear as tool parameters or in source files.
No error recovery guidance. All try-catch blocks console.error() and return undefined. LLMs receive no indication of what went wrong (authentication failed? network error? invalid symbol?) or what to do next.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 34 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 37 | - | v1 |
Tool descriptions are too short (<20 chars) for effective LLM selection. 'Get current open positions' and 'Get current stock holdings' lack context on when to use each, what data structure they return, or prerequisites.
Destructive operations (sellStock, buyStock) have no confirmation step or dry-run capability. Agents can accidentally execute real trades. No audit trail or recovery mechanism.
The 'add' tool is unrelated to trading functionality and pollutes the tool namespace. It appears to be a leftover test tool. Remove it or move to a separate testing server.
No permission gating. Any agent with access to this server can execute trades. There is no validation of agent identity, user authorization, or least-privilege scoping.
Quantity parameter accepts any number (z.number()) with no minimum or maximum. LLMs could pass negative quantities, zero, or absurdly large orders causing API errors or financial loss.
Return values do not include chaining IDs. After buyStock succeeds, the agent receives only 'Stock bought successfully' (text string). No order_id, transaction_id, or reference to track the trade.