Auditra WordPress MCP
Auditra is a free WordPress plugin that lets an AI assistant like Claude read everything about your site's plugins — installed, deleted, and forgotten — and answer your questions about them in plain English.
README
Auditra
A WordPress plugin that turns the site it is installed on into a read-only MCP server, so any MCP-capable AI client (Claude custom connectors, and others) can inspect and reason about that site's plugin estate in plain language.
Ask "which of my plugins are vulnerable?", "what's bloating my options table?", or "what did deleted plugins leave behind?" — and get answers grounded in the actual state of the site.
How it works
Three layers inside one plugin, one REST route, no runtime dependencies:
- Transport —
POST /wp-json/auditra/v1/mcp/{token}speaking MCP over JSON-RPC 2.0 (plain JSON responses, stateless). Both protocol generations are served, decided per request: revision2026-07-28(per-request metadata,server/discover,Mcp-*header validation) and theinitialize-handshake revisions2025-11-25/2025-06-18/2025-03-26. Token-in-path auth behind a swappable interface, Origin validation, rate limiting, failed-auth logging. - Collectors — read WordPress directly: inventory, site context, autoloaded options, cron, database tables, content-feature usage, and a three-tier attribution engine that maps options/tables/hooks back to owning plugins with explicit confidence levels (
high= curated,medium= derived, or visibly unattributed). - Enrichment — two external services, wordpress.org and WPVulnerability, both keyless, both cached, both degrading silently with per-source coverage reporting and progressive backoff (15 min → 24 h) on failures. Support lifecycle dates for PHP, MySQL and MariaDB are not fetched at all: they ship in
includes/data/lifecycle.json, compiled from each vendor's published policy.
Nine tools; every response carries _meta (totals, truncation, sources unavailable, timestamps) and stays within a 20 KB budget. The server reports facts, never verdicts — scoring and advice are deliberately absent, because that's the AI client's job. The full design record lives in docs/DECISIONS.md.
Connect it to an AI client
- Tools → Auditra in wp-admin: enable the endpoint, generate a token, copy the connection URL.
- In Claude: Settings → Connectors → Add custom connector, paste the URL (must be
https://). - Ask questions. Start with "what can you tell me about this site's plugins?"
The site must be publicly reachable over HTTPS with pretty permalinks enabled. Clients on revision 2026-07-28 get a 24-hour freshness hint on tools/list, so a tool added by a plugin update appears within a day on its own; older revisions have no expiry signal, so reconnect the connector after updating to pick up new tools.
Contributing
The easiest genuinely useful contribution is a pull request against
includes/data/prefix-overrides.json — the curated map from plugin slugs to the option/table prefixes they actually use (Contact Form 7 → wpcf7_). Every entry improves attribution accuracy for every user. See CONTRIBUTING.md.
Development
composer install # dev-only: PHPCS + WordPress Coding Standards (pinned)
composer lint
php tests/mcp-client.php https://example.com/wp-json/auditra/v1/mcp/{token}
tests/mcp-client.php is the CLI harness that exercises the endpoint without an AI client in the loop; tests/seed-conditions.sh builds a deliberately messy test site (and teardown-conditions.sh reverses it exactly). CI runs a PHP lint matrix (7.4–8.4), PHPCS, and a grep gate that fails the build if any write operation ever appears in the plugin code.
Security
See SECURITY.md for the supported versions and how to report a vulnerability privately.
License
GPL-2.0-or-later.
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。