Web & Academic Search MCPs
Relevant source files
The following files were used as context for generating this wiki page:The AI Dev Atelier project utilizes a suite of Multi-Capability Providers (MCPs) to equip AI agents with powerful web and academic search capabilities. These MCPs are external services integrated into the agent's workflow, primarily through the search and research skills. They provide the foundational tools for agents to gather information, look up documentation, conduct literature reviews, and perform evidence-based analysis.
The core purpose of these MCPs is to ground agent responses and decisions in external, verifiable data. The search skill leverages MCPs for quick lookups and general web queries, while the research skill employs a more structured, multi-MCP workflow to conduct comprehensive academic research, culminating in synthesized reports and evidence cards. This system enables agents to move beyond their pre-trained knowledge and interact with real-time, domain-specific information.
Sources: skills/search/SKILL.md, skills/research/SKILL.md, AGENTS.md:73-86, README.md:36-50
MCP Configuration
All MCP servers are defined in the mcp.json file. This configuration file specifies the server endpoints and is used by the install.sh script to set up the MCPs for agent use. Agents require specific API keys, set as environment variables, to access the functionality of most search and research MCPs.
Sources: AGENTS.md:21, README.md:21
MCP Server Definitions
The table below summarizes the primary search and research MCPs, their purpose, and the required configuration.
| MCP Server | Purpose | Required Key / Env Var |
|---|---|---|
tavily-remote-mcp | Web search and research | TAVILY_API_KEY |
openalex-research | Academic papers and citations | OPENALEX_EMAIL |
pdf-reader | PDF text extraction | None |
paper-search | Multi-platform paper search | Optional keys |
context7 | Library documentation search | CONTEXT7_API_KEY (if used) |
grep | GitHub code search | None |
zai-zread | GitHub repo semantic search (issues, PRs, docs) | Z_AI_API_KEY |
zai-vision | Image/video analysis, UI code gen, diagrams | Z_AI_API_KEY |
Sources: AGENTS.md:73-86, README.md:36-50
Core Search & Research MCPs
The search and research functionalities are powered by a combination of specialized MCPs, each serving a distinct role in the information retrieval process.
Tavily (Web Search)
Tavily is the primary MCP for general web search. It is the cornerstone of the search skill, used for looking up documentation, code examples, API references, and troubleshooting guides. The research skill also uses Tavily in parallel with academic sources to find non-academic materials like blog posts and case studies, which are considered valuable for SOTA (State-of-the-Art) context but are weighted lower than peer-reviewed papers.
Sources: skills/search/SKILL.md, skills/research/SKILL.md, skills/research/references/guides/guide-workflow.md:63-70, AGENTS.md:75
OpenAlex (Academic Research)
OpenAlex is the core MCP for academic discovery within the research skill. It provides tools to find foundational, highly-cited papers as well as recent, cutting-edge research. Agents use it to build a comprehensive understanding of a topic grounded in academic literature.
Key OpenAlex tools used by the research skill include:
get_top_cited_works: To find influential, foundational papers.search_works: To find recent or SOTA research, often filtered by publication year.get_citation_network: To explore related works by following citation links.
Sources: skills/research/SKILL.md:60-62, skills/research/references/guides/guide-workflow.md:46-58, AGENTS.md:78
Paper-search & PDF-Reader
These two MCPs work in tandem within the research skill.
paper-search: An optional MCP used to find and download papers from multiple platforms (e.g., arXiv, PubMed).pdf-reader: A crucial tool for extracting text content from downloaded PDF files. This allows the agent to "read" the paper and synthesize its contents into evidence cards.
Sources: skills/research/SKILL.md:63-64, AGENTS.md:79-80, README.md:46-47
Agent Interaction & Workflow
Agents do not call MCPs directly but interact with them through structured skills. This provides a layer of abstraction and enforces best practices for information retrieval.
search Skill Workflow
The search skill provides a straightforward workflow for web queries. It is designed for quick lookups and fact-finding missions.
The diagram below illustrates the simple flow of the search skill.
Sources: skills/search/SKILL.md, skills/README.md:44-46
research Skill Workflow
The research skill orchestrates a more complex, multi-step workflow involving several MCPs. The goal is not just to find information but to synthesize it into a comprehensive report with at least five evidence cards that cover diverse and potentially conflicting approaches.
This workflow diagram shows the comprehensive research process.
A critical part of this workflow is the "Write-First Pattern," where an agent must write or update an evidence card and save its state after reading a maximum of two sources. This prevents context overload and ensures a persistent knowledge store.
Sources: skills/research/SKILL.md:23-47, skills/research/references/guides/guide-workflow.md:1-25, 31-41
MCP Integration Summary
This diagram provides a high-level overview of how skills act as an intermediary between an AI agent and the various search-related MCPs.
Sources: skills/README.md:10-21, AGENTS.md:31-41, README.md:26-34
Conclusion
The Web & Academic Search MCPs are a cornerstone of the AI Dev Atelier project, providing the essential capabilities for agents to perform informed, evidence-based tasks. By integrating specialized services like Tavily for web search and OpenAlex for academic research through the search and research skills, the system ensures that agents can access and synthesize a wide range of external information. The structured workflows, particularly the comprehensive process of the research skill, transform simple data retrieval into a robust system for analysis and knowledge creation.