MCQuest MCP
A read-only MCP server that gives AI coding agents structured access to a project's source code, architecture, documentation, and Git context through 16 tools for searching, reading, and comparing evidence without modifying files.
README
MCQuest MCP — Read-Only Project Intelligence Server
A reusable, read-only MCP server that gives AI coding agents structured access to a software project's source code and documentation.
Built for and battle-tested on the MCQuest codebase, the server is intentionally maintained as a standalone developer tool outside application repositories so it can be pointed at any repository.
It is an evidence layer, not a decision engine: it reports what is actually in a project, and the AI coding agent performs the reasoning.
Features
Provides 16 read-only tools organized into categories:
Project Structure
| Tool | Purpose |
|---|---|
mcquest_project_info |
Compact project overview with structure and key files |
mcquest_project_context |
High-level understanding: tech stack, directories, source areas |
mcquest_list_files |
Recursively list project files with glob pattern filtering |
Source Code Analysis
| Tool | Purpose |
|---|---|
mcquest_read_file |
Read bounded file sections with line numbers |
mcquest_search |
Regex-based source code search with context |
mcquest_find_files |
Find files by filename (case-insensitive) |
mcquest_find_imports |
Find ES module imports referencing a target |
mcquest_find_usages |
Find symbol references with word-boundary matching |
mcquest_pattern_audit |
Predefined responsive/layout pattern audit |
Documentation Intelligence
| Tool | Purpose |
|---|---|
mcquest_list_docs |
List Markdown documentation files |
mcquest_read_doc |
Read Markdown docs with line numbers |
mcquest_search_docs |
Regex/text search across docs |
mcquest_phase_context |
Locate phase docs (supports query and phase modes) |
mcquest_compare_phase |
Compare documentation between two phases |
Cross-Evidence & Git
| Tool | Purpose |
|---|---|
mcquest_find_evidence |
Search code + docs in one call, grouped by evidence type |
mcquest_git_context |
Read-only Git: branch, status, recent commits, changes |
The server does not modify the target project.
Quick Start
Requires Python 3.10+ and uv.
# 1. Get the code
# (or clone your fork / place it somewhere outside your app repos)
cd mcquest-mcp
# 2. Install dependencies (runtime dep is only the MCP SDK;
# pytest is installed as a dev-only dependency)
uv sync
# 3. Run it, pointing the server at the repository you want to inspect
# Windows (PowerShell):
uv run mcquest-mcp --project "D:\path\to\your\project"
# macOS / Linux:
uv run mcquest-mcp --project /path/to/your/project
Important: you must tell the server which repository to inspect. Either pass
--project <PATH>(absolute or relative), or set theMCQUEST_PROJECT_ROOTenvironment variable. If neither is supplied, startup fails with a clear error — the server never silently assumes a repository.--projecttakes precedence over the environment variable.
The same installed server is safely reused across repositories by changing the
--project argument, e.g.:
uv run mcquest-mcp --project "D:\DeveloperTools\mcquest-mcp" # this repo
uv run mcquest-mcp --project "D:\Some\Other\Repository" # anything else
Configuration
| Setting | Default | Purpose |
|---|---|---|
--project / MCQUEST_PROJECT_ROOT |
(none — startup fails if unset) | Absolute or relative path to the project root all tools are confined to |
- Path containment. Every tool resolves its inputs against
MCQUEST_PROJECT_ROOTand rejects path traversal (../, absolute escapes) outside that boundary. - Default search paths. Several tools default to
frontend/src(source) anddocs(documentation). Passpath=/docs_path=explicitly for other layouts. - Phase docs.
mcquest_phase_contextandmcquest_compare_phaseassume docs are named likedocs/phase-<N>...mdunder the documentation path.
Registering with an MCP client
Wire the mcquest-mcp console script into your MCP client's server configuration.
Example (JSON, e.g. Cline/Claude local config):
{
"mcpServers": {
"mcquest-mcp": {
"command": "uv",
"args": ["run", "mcquest-mcp", "--project", "C:/path/to/your/project"]
}
}
}
Or run the MCP Inspector during development:
uv run python -m mcp dev dev_server.py
Documentation Awareness
The documentation tools give the AI agent read-only access to project Markdown
documentation (e.g. docs/, memory-bank/), including previous audit and
implementation phases.
mcquest_list_docs— list Markdown files under a project-relative path.mcquest_read_doc— read a Markdown file with exact line numbers.mcquest_search_docs— regex/text search across Markdown docs, returning file path, line number, matching line, and surrounding context.mcquest_phase_context— Two modes:querymode (free-text) andphasemode (structured lookup, e.g.phase="61"). Phase mode groups results by document type.
These tools preserve the original Markdown evidence and line references. They do not summarize, rewrite, or interpret the documentation; they return the raw evidence so the AI agent can reason about it.
Safety
This MCP is intentionally read-only.
It does not provide:
- file editing
- file creation
- file deletion
- shell execution
- arbitrary code execution
- package installation
- Git mutation
- commits
- pushes
- deployment
The server provides evidence; the AI coding agent performs the reasoning. It never
claims runtime verification — static/documentation evidence is always labeled as such
(e.g. RUNTIME VERIFICATION: NOT PERFORMED where applicable).
Recommended Cline Usage Flow
mcquest_project_context— First call for project orientationmcquest_phase_context(phase="N")— Understand a specific phasemcquest_find_evidence(query="...")— Cross-search code + docsmcquest_read_file/mcquest_read_doc— Read specific filesmcquest_git_context— Check recent changes before investigationmcquest_compare_phase(from_phase="61", to_phase="62")— Compare phase docs
Development & Testing
-
uv syncinstalls runtime + dev dependencies. -
Run the regression suite with:
uv run pytest -qThe suite guards: the exact 16-tool registration, path-security/traversal, the read-only guarantee, output bounds, evidence-neutral
compare_phaseoutput, and thefind_evidence(scope="phase")behavior. -
Runtime dependency is only
mcp[cli]>=2,<3;pytestis a dev-only dependency.
Known Legacy Defaults
The tool names retain the mcquest_ prefix and several defaults assume an
MCQuest-style layout (e.g. frontend/src, docs, specific config filenames in
mcquest_project_info / mcquest_project_context, responsive-pattern categories in
mcquest_pattern_audit). These are intentionally retained for backward
compatibility while the server is progressively generalized. The project root is no
longer hard-coded: it comes from --project or MCQUEST_PROJECT_ROOT, and startup
fails if neither is supplied. Git tools are limited to read-only inspection
commands.
Version History
- v0.3.0 — Evidence-neutral
compare_phasevocabulary (ADDED/REMOVED/COMMON,RUNTIME VERIFICATION: NOT PERFORMED), fixedfind_evidence(scope="phase"), added a dev-only pytest regression suite. - v0.2.0 — Added
project_context,find_evidence,git_context,compare_phase. Improved all tool descriptions. Addedphaseparameter tophase_contextwith structured lookup grouped by document type. - v0.1.0 — Initial release with 12 read-only tools.
Requirements
- Windows, macOS, or Linux
- Python 3.10+
uv- MCP Python SDK
- Node.js only if using MCP Inspector
Install
Clone or place the server somewhere outside your application repositories, then see Quick Start.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。