Loading ToolBench

Reading the current corpus…

Deep-Research-MCP-Server MCP server · D · ToolBench

Deep-Research-MCP-Server

github.com/lkk989/Deep-Research-MCP-Server

An AI-powered research assistant that performs iterative, deep research on any topic by combining search engines, web scraping, and Gemini large language models. Available as a Model Context Protocol (MCP) tool for seamless integration with AI agents.

<=2025-11-25STDIO1 tools

51/100

Definition Quality52

Protocol Readiness47

SupportabilityNot measured-

Spec posture

Inferred effective spec <=2025-11-25·current baseline 2026-07-28

Static source inference · medium confidence · detected: Logging

Deprecated protocol patterns detected

Summary

Single tool 'deepResearch.run' has a complete input schema with proper Zod validation and type constraints (query, depth 1-5, breadth 1-5, existingLearnings array, goal, flags object). Description is present (71 chars) but generic, does not explain WHEN to use this vs. alternatives, prerequisites, or what 'Gemini-only' means to an agent. Output schema is partially documented in code (MCPResearchResult interface with content/metadata/stats) but NOT exposed in tool registration, LLMs cannot see the return structure. No error handling guidance; catch block returns generic error text without recovery hints. Tool name 'deepResearch.run' uses 'run' (generic verb, not action-specific like 'search_' or 'research_'). Parameters lack descriptions in the schema registration, only present in Zod chain but not surfaced to MCP clients. No pagination, rate limiting, or result limits documented.

Tools (1)

deepResearch.runread onlyauthsource verified52/100

Gemini-only deep research pipeline (Google Search grounding + URL context).

Issues (5)

HIGHtool-naming ↗

Tool name 'deepResearch.run' uses generic verb 'run' instead of action-specific verb (e.g., 'search_research', 'conduct_research'). LLMs cannot infer intent from 'run' alone.

deepResearch.run

HIGHresponse-shaper ↗

Output schema (MCPResearchResult with content/metadata/stats/learnings/visitedUrls) is defined in TypeScript but NOT documented in tool registration. LLMs cannot see return structure and cannot plan downstream operations.

deepResearch.run

HIGHtool-description ↗

Tool description (71 chars) is too generic: 'Gemini-only deep research pipeline (Google Search grounding + URL context).' Does not explain WHEN to use it, what 'Gemini-only' means to agent, or prerequisites (API key, rate limits).

deepResearch.run

HIGHrecovery-guide ↗

Error handling returns generic error text without recovery guidance. Catch block: 'Error during deep research: {message}' does not tell LLM what to do next (retry? ask user? fatal?).

deepResearch.run

Recommendations

Rename tool from 'deepResearch.run' to 'search_research' or 'conduct_research' to use action-specific verbs. This helps LLMs infer intent from the name alone.
Expand tool description to 150 - 250 chars: explain WHEN to use (complex multi-faceted queries), prerequisites (GEMINI_API_KEY required), and what 'Gemini-only' means (uses Google Search grounding via Gemini API, not standalone search).
Document output schema in tool registration via a 'returns' field or inline comment. Include field types and meanings: content (array of text blocks), metadata.learnings (array of key findings), metadata.visitedUrls (array of sources), metadata.stats (counts).
Add error classification to catch block: return structured error with category (retryable|user_fixable|fatal) and recovery hint. E.g., 'API rate limit exceeded. Retry in 60 seconds.' or 'Invalid query format. Ensure query is a non-empty string.'
Document result limits in description and enforce in code: cap learnings to 50 and visitedUrls to 100. Return pagination cursor if more results exist. State: 'Returns up to 50 learnings and 100 sources; use cursor for additional results.'

Spec posture evidence

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

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

Score history

Overall score trend

First recorded score · v2 rubric

51/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	51	<=2025-11-25	v2