Hardcoded Algolia API Credentials in Source Code
The apify-mcp-server contains hardcoded Algolia API credentials (appId and apiKey) embedded in the source code at src/const.ts. These credentials are used for the docu...
Read advisoryTouchstone research powers the deeper checks behind public registry scores and publisher reviews.
The apify-mcp-server contains hardcoded Algolia API credentials (appId and apiKey) embedded in the source code at src/const.ts. These credentials are used for the docu...
Read advisorymcp-grafana trusts the inbound X-Grafana-URL header in HTTP transports and combines it with environment fallback for credentials. If the request omits X-Grafana-Servic...
Read advisorymcp-grafana's HTTP transport initializes Grafana client context per request using header/env inputs, but does not implement a mandatory authentication or authorization...
Read advisoryTask.output_file is validated only once, before {...} placeholders are interpolated. If an application uses a task configuration like output_file="{path}" or output_fi...
Read advisoryThe HTTP transport path does not implement authentication middleware or API-key/session authorization checks. Requests without Origin bypass CORS checks by design, so...
Read advisoryThe HTTP middleware trusts client-controlled TFE_ADDRESS values from headers or query parameters and propagates them into per-session client creation. The TFE client i...
Read advisorygithub-mcp-server applies sanitize.Sanitize() inconsistently. It is applied in the GraphQL fragment path (fragmentToMinimalIssue:391-392) for issue titles and bodies....
Read advisorygithub-mcp-server exposes several destructive write tools that execute immediately on the first call from the LLM with no server-side confirmation step: - merge_pull_r...
Read advisoryThe amazon-bedrock-agentcore-mcp-server in the awslabs/mcp monorepo includes a browser_evaluate MCP tool that passes an LLM-controlled expression parameter directly to...
Read advisoryThe iam-mcp-server in the awslabs/mcp monorepo exposes IAM write tools that accept arbitrary policy content with no validation beyond JSON syntax checking. The server...
Read advisoryThe lambda-tool-mcp-server in the awslabs/mcp monorepo uses a fail-open scope model: when neither FUNCTION_PREFIX nor FUNCTION_LIST environment variables are set (the...
Read advisoryuse_sentry creates an embedded AI agent that must use tools and "return tool results directly without interpreting them." That agent receives raw tool output from Sent...
Read advisoryThe MCP server's event formatters do not apply a sensitive-data boundary before passing Sentry data into LLM-facing outputs. The code renders stack-frame local variabl...
Read advisoryThe gateway.ts HTTP client in @unusualwhales/mcp propagates the full, unredacted HTTP response body from the Unusual Whales API back to MCP tool callers when errors oc...
Read advisoryThe create-my-tools-profile skill asks the user (or any LLM context in scope) for a free-form role description and writes it directly into a persistent AI rules file (...
Read advisoryThe
Read advisoryThe @unusualwhales/mcp server registers 25+ MCP prompts (workflow templates) that compose LLM instructions by directly interpolating user-supplied arguments into templ...
Read advisoryThe browser_navigate MCP tool in @playwright/mcp passes user-supplied URLs through checkUrlAllowed() (context.ts) which only blocks the file: protocol. The default con...
Read advisoryThe type() method in mobile-use performs incomplete shell escaping when sending text to Android devices via ADB. While it escapes whitespace and double quotes, it does...
Read advisoryThe command validation in desktop-commander uses a blocklist approach that can be bypassed using $() command substitution or backtick substitution. For example, $(curl...
Read advisoryThe node:local feature in desktop-commander allows arbitrary code execution by writing user-provided JavaScript to a temporary .mjs file and executing it with Node.js....
Read advisorygoogle-adk's DatabaseSessionService supports both a v0 schema (ADK 1.19.0–1.21.0) and a v1 schema. The service detects the existing schema version on startup and uses...
Read advisoryThe navigate MCP tool (mcp-server.js:41-48), HTTP /navigate endpoint (server.js:44-49), and batch action handler (mcp-server.js:339-342) all pass user-supplied URLs di...
Read advisoryThe Express HTTP server in server.js listens on port 3033 (configurable via PORT env var) bound to all network interfaces (0.0.0.0 — Express default when no host is sp...
Read advisorychrome-local-mcp exposes an eval MCP tool (mcp-server.js:198-205) and an HTTP /eval endpoint (server.js:117-122) that pass user-supplied JavaScript strings directly to...
Read advisory`@osematouati/notion-mcp-server` (versions 1.8.1–1.8.2) is a third-party npm republication of the official Notion MCP server, published by a single maintainer (`osemat...
Read advisoryAll 7 published versions (0.0.1 through 0.0.8) of `@gongrzhe/server-gmail-mcp` on npm lack provenance attestation. The npm registry metadata contains `dist.signatures`...
Read advisory63 checks feed 12 public score categories.
OAuth 2.1 implementation, PKCE enforcement, token storage, HTTPS enforcement, scope analysis, session management, RFC 8707 compliance.
Prompt injection in tool schemas, parameter poisoning, obfuscated payloads, tool shadowing, rug pull detection, dangerous capability combinations.
SSRF via tool parameters, cloud metadata endpoint access, command injection, SQL injection, path traversal, URL scheme validation.
Credential patterns in schemas, PII exposure, secrets in error messages, sensitive data in URL parameters, cross-server data leakage.
npm provenance verification, known CVE matching, typosquat detection, maintainer reputation, source repo verification, abandonment detection.
Network binding audit, TLS enforcement, rate limiting, CORS configuration, error handling, security headers, DNS rebinding protection.
Guardrail bypass patterns, response size limits, timeout enforcement, and concurrency safety signals that feed buyer confidence and deeper review.
Google A2A Agent Card security: prompt injection in descriptions, obfuscated content, identity spoofing, HTTPS enforcement, capability over-privilege.
Demographic signal detection in tool parameters, differential treatment risk assessment, and data governance gap analysis. Maps to EU AI Act Article 10.