MCP Services Overview
Relevant source files
The following files were used as context for generating this wiki page:
The AI Dev Atelier project integrates a suite of external services through the Model Context Protocol (MCP) to extend the capabilities of AI agents. These services provide functionalities like web search, academic research, code search, and documentation lookup, which are leveraged by various skills within the repository. The central configuration for these services is managed in the mcp.json file.
This system allows skills such as search and research to access external information and tools, making the AI agents more powerful and context-aware. The install.sh script automates the configuration of these MCP servers for agents like OpenCode.
Sources: mcp.json, README.md:27-46, INSTALL.md:168-171
Architecture and Configuration
MCP services act as a bridge between the AI agent's skills and external APIs or local commands. When a skill needs to perform an action like searching the web, it calls the corresponding MCP server, which handles the request and returns the result.
The diagram below illustrates the relationship between an AI Agent, the skills it uses, and the underlying MCP services.
This diagram shows how the AI Agent uses skills, which in turn delegate specific tasks to the appropriate MCP servers.
Sources: mcp.json, skills/README.md:11-23
Configuration and Installation
MCP servers are defined in mcp.json. The install.sh script reads this file and automatically configures the MCPs for OpenCode by updating ~/.opencode/opencode.json. This process requires the jq command-line tool to be installed. For other agents, the mcp.json file can be copied and configured manually. API keys are sourced from a .env file during installation.
Sources: INSTALL.md:168-185, AGENTS.md:78-81
MCP Services Summary
The project defines several MCP servers, each with a specific purpose.
| MCP Server | Purpose | Associated Skills | API Key / Env Var |
|---|---|---|---|
tavily-remote-mcp | Web search and research. | search | TAVILY_API_KEY |
context7 | Library documentation search. | search | CONTEXT7_API_KEY (optional) |
grep | Search across public GitHub repositories. | search | None |
openalex-research | Academic paper discovery and citation analysis. | research | OPENALEX_EMAIL |
pdf-reader | PDF text extraction. | research | None |
paper-search | Multi-platform academic paper search and download. | research | Optional keys |
zai-zread | GitHub repo semantic search (issues, PRs, docs). | search, research | Z_AI_API_KEY |
zai-vision | Image/video analysis, UI code generation, diagram understanding. | ui-animation, code-review, docs-write, research | Z_AI_API_KEY |
graphite | Stacked PR management. | use-graphite | None (requires Graphite CLI) |
Sources: README.md:27-46, mcp.json, AGENTS.md:83-98
Detailed MCP Definitions
This section provides details for each MCP server as defined in mcp.json.
tavily-remote-mcp
Used by the search skill for general web searches. It connects to a remote URL and requires an API key.
"tavily-remote-mcp": {
"_description": "Tavily MCP - Web search and research. Required for 'search' skill.",
"url": "https://mcp.tavily.com/mcp",
"headers": {
"Authorization": "Bearer TAVILY_API_KEY"
}
}
Sources: mcp.json:5-11
context7
Provides developer documentation search capabilities for the search skill.
"context7": {
"_description": "Context7 MCP - Developer documentation search. Required for 'search' skill.",
"url": "https://mcp.context7.com/mcp",
"headers": {
"CONTEXT7_API_KEY": "CONTEXT7_API_KEY"
}
}
Sources: mcp.json:12-18
grep
Enables code search across a million public GitHub repositories for the search skill. It does not require an API key.
"grep": {
"_description": "Grep MCP - Search across a million public GitHub repositories. Required for 'search' skill.",
"url": "https://mcp.grep.app"
}
Sources: mcp.json:19-22
openalex-research
A command-based MCP used by the research skill for academic paper discovery. It requires an email address for the OpenAlex API.
"openalex-research": {
"_description": "OpenAlex MCP - Academic paper discovery and citation analysis. Required for 'research' skill.",
"command": "npx",
"args": ["-y", "openalex-research-mcp"],
"env": {
"OPENALEX_EMAIL": "you@example.com"
}
}
Sources: mcp.json:23-30
pdf-reader
A command-based MCP for extracting text from PDF files, primarily used by the research skill.
"pdf-reader": {
"_description": "PDF Reader MCP - PDF text extraction. Required for 'research' skill.",
"command": "npx",
"args": ["-y", "@sylphx/pdf-reader-mcp"]
}
Sources: mcp.json:31-35
zai-zread & zai-vision
These MCPs from Z.AI provide advanced semantic search and vision capabilities. They require a Z_AI_API_KEY. The configuration includes notes on managing API quota by disabling certain tools that overlap with native agent capabilities.
For example, AGENTS.md suggests disabling zai-zread_read_file and using webfetch instead to save on quota.
"zai-zread": {
"_description": "Z.AI Zread MCP - GitHub repo semantic search (issues, PRs, docs). Optional for 'search' and 'research' skills.",
"_quota": "1,000 calls/month shared pool (Pro plan). Only search_doc is unique.",
"_skills": ["search", "research"],
"_enabledTools": {
"search_doc": "Semantic search across repo issues, PRs, docs - UNIQUE capability"
},
"_disabledTools": {
"_comment": "Disable in agent config to save quota - use free alternatives",
"read_file": "Use webfetch on https://raw.githubusercontent.com/{owner}/{repo}/{branch}/{path}",
"get_repo_structure": "Use gh api repos/{owner}/{repo}/git/trees/{branch}?recursive=1"
},
"url": "https://api.z.ai/api/mcp/zread/mcp",
"headers": {
"Authorization": "Bearer Z_AI_API_KEY"
}
}
Sources: mcp.json:42-61, AGENTS.md:100-113
graphite
This MCP is used by the use-graphite skill to manage stacked pull requests. It is a command-based MCP that requires the Graphite CLI to be installed.
"graphite": {
"_description": "Graphite MCP - Stacked PR management. Required for 'use-graphite' skill.",
"_prerequisite": "Graphite CLI v1.6.7+ must be installed (npm install -g @withgraphite/graphite-cli)",
"_skills": ["use-graphite"],
"command": "gt",
"args": ["mcp"]
}
Sources: mcp.json:86-92
Conclusion
The Model Context Protocol (MCP) is a core architectural component in AI Dev Atelier, providing a standardized way for skills to interface with a wide range of external services. This modular approach enhances agent capabilities for complex tasks like research, code search, and documentation analysis. The system is designed for easy configuration and extension through the mcp.json file and automated setup scripts.
Sources: mcp.json, README.md, INSTALL.md