getdebug MCP
Lets MCP-compatible AI clients read your getdebug projects, findings, and proposed fixes using your existing bearer token.
README
@getdebug/mcp
Model Context Protocol server for getdebug. Lets Claude, Cursor, and any MCP-compatible AI client read your projects, findings, and proposed fixes through the same bearer token getdebug login already creates.
Tools (v0.2)
list_projects— projects in your active org, with last-run status + finding counts.list_findings— findings for a project, filterable by severity, with id / file / line / CWE / OWASP.get_finding— full details of one finding: explanation, snippet, CWE/OWASP refs, and the proposed-fix diff when one exists.list_fixes— proposed/applied fixes (optionally scoped to a project or status), with finding link + PR URL.start_scan— enqueue a fresh hosted analyze run on a project. Findings appear in ~1-2 min; re-calllist_findingsto fetch.
apply_fix lands in a future release. Today: when get_finding shows a proposed-fix diff, applying it still happens via the PR review flow on the dashboard.
Auth
The server reads ~/.getdebug/config.json, the same file getdebug login writes. You must:
- Install the CLI:
npm i -g @getdebug/cli - Run
getdebug loginonce. - The config file must be
chmod 600— the server refuses to load with looser perms, matching the CLI.
No new credentials, no extra OAuth, no MCP_TOKEN env var to manage.
Setup — Claude Desktop
Add this to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on your OS:
{
"mcpServers": {
"getdebug": {
"command": "npx",
"args": ["-y", "@getdebug/mcp"]
}
}
}
Restart Claude Desktop. The three tools appear under the 🔧 menu in any chat.
Setup — Cursor
Cursor's MCP config lives at ~/.cursor/mcp.json (or per-workspace at .cursor/mcp.json):
{
"mcpServers": {
"getdebug": {
"command": "npx",
"args": ["-y", "@getdebug/mcp"]
}
}
}
Setup — any other MCP client
Transport is stdio. Spawn getdebug-mcp (or npx @getdebug/mcp); the server speaks MCP over the child's stdin/stdout. Stderr is for logs only.
Local development
pnpm install --filter @getdebug/mcp...
cd mcp
pnpm dev # tsx, hot-reload
pnpm typecheck
pnpm build && node dist/index.js
Override the API base for staging or local dev:
GETDEBUG_API_URL=http://localhost:3001 node dist/index.js
(The CLI honors the same env var — see cli/internal/cmd/login.go.)
What the agent will see
Example: an agent asks "any high-severity findings in my debug project?" The tool flow:
list_projects→ picks the project id matching "debug".list_findings({projectId, severity: "high", limit: 25})→ returns a list of HIGH findings with file paths + line numbers.get_finding({findingId})for any the agent wants to drill into → returns explanation + snippet + proposed-fix diff.
The agent can now reference your actual security findings by file and line while answering questions or writing code — without you copying anything in.
Privacy
This server is a thin client. It never touches your source files directly; it only reads what's already in your getdebug org via the same API the CLI and dashboard use. Run getdebug login --logout (or just delete ~/.getdebug/config.json) to revoke access.
License
MIT.
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。