Loading ToolBench

Reading the current corpus…

AI-Agent-Projects MCP server · F · ToolBench

AI-Agent-Projects

Collection of AI agent projects using LangChain, LangGraph, and MCP (Model Context Protocol) for building multi-tool agents with various integrations including Gmail, Google Sheets, Airbnb, Yahoo Finance, and custom tools.

2026-07-28+STDIO2 tools

43/100

Definition Quality61

Protocol Readiness15

Supportability31

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · detected: Dynamic Client Registration

Deprecated protocol patterns detected

Summary

This server exposes only 2 tools with minimal definition quality. The tools are directly borrowed from LangChain's @tool decorator and lack the structured JSON Schema rigor required for reliable agent invocation. The server itself is not an MCP server, it's a LangChain/FastAPI application that references MCP adapters but does not expose an MCP interface. The two tools (web_search, get_weather) are defined in scripts/base_tools.py using LangChain's @tool decorator, not MCP ToolDefinition objects.

Tools (2)

get_weatherread onlyauthsource verified65/100

Get current weather for a location using WeatherAPI.com. Use for queries about weather, temperature, or conditions in any city. Examples: 'weather in Paris', 'temperature in Tokyo', 'is it raining in London'

web_searchread onlysource verified57/100

Perform a live web search using Ollama Cloud Web Search API for real-time information and news.

Issues (6)

CRITICALtool ↗

No output schema documented for either tool. LLMs cannot plan downstream operations or extract structured data without knowing what fields to expect.

web_searchget_weather

HIGHconstrained-input ↗

Input parameters lack formal type constraints and constraints. web_search accepts 'query' (string) with no length limits, max_results, or language hints. get_weather accepts 'location' with no validation of expected format (city name vs lat/lon).

web_searchget_weather

HIGHrecovery-guide ↗

No error handling guidance. Tools call external APIs (Ollama Web Search, WeatherAPI) without documented fallback behavior or recovery instructions for LLMs. If Ollama returns malformed JSON or WeatherAPI rate-limits, LLM has no next-step guidance.

web_searchget_weather

MEDIUMtool-annotations ↗

No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Both tools are read-only but lack formal marking. This increases LLM uncertainty about retry safety.

Recommendations

Document output schema for web_search: describe the structure of response.results (list of dict with fields: title, url, snippet, source, etc.). Specify max result count (currently 2) and whether results are ranked by relevance.
Document output schema for get_weather: describe the WeatherAPI.com response structure (current, location, last_updated fields and their types). Truncate irrelevant metadata (audit fields, internal timestamps) to reduce token waste.
Add constraints to web_search 'query' parameter: max 500 characters, required non-empty string. Add optional 'max_results' parameter (default 2, range 1-10) to let agents tune result count.
Add constraints to get_weather 'location' parameter: accept city name (e.g. 'Paris', 'New York') or 'city, country' (e.g. 'London, UK'). Document that coordinates (lat/lon) are NOT supported. Suggest search_cities tool for ambiguous location names.
Add error handling guidance: 'If location not found, call search_cities() to find the exact spelling or use 'city, country' format. If WeatherAPI rate-limits (too many calls), retry after 60 seconds.'
Mark both tools with readOnlyHint=true in tool definition to signal to LLMs that these are safe to retry.
Refactor get_weather to accept API key as server-side secret only. Remove os.getenv() call from tool function, inject key at server initialization time or via environment configuration that never appears in logs.
If converting this to a true MCP server, explicitly register tools via ToolDefinition (not just @tool decorator). Include structured JSON Schema for inputs and document outputs in the definition.

Score history

Overall score trend

↑ 12 points across a rubric change (v1 → v2)

43/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	F	43	2026-07-28+	v2
2026-03-09	F	31	-	v1