Stop Searching. Start Trusting.

The curated directory of MCP servers, vetted for security, efficiency, and quality.

Browse DirectoryWhy Us?

Tired of the MCP "Marketplace" Chaos?

We built MCPScout.ai to solve the ecosystems biggest pain points.

No Insecure Dumps

We manually analyze every server for basic security flaws.

Easy Setup

Our gotcha notes warn you about complex setups.

Avoid "Token Hogs"

We estimate token costs for cost-effective agents.

Products, Not Demos

We filter out "Hello World" demos.

CATEGORIES:
SORT:

Vetted Servers(9120)

0
0
Low Cost
ArdaGurcan icon

cloudlab-mcp

by ArdaGurcan

Sec9

Enables AI assistants to directly manage and interact with CloudLab/Emulab experiments for provisioning, monitoring, and controlling compute resources.

Setup Requirements

  • ⚠️Requires Node.js 18+.
  • ⚠️Requires a CloudLab account with API access and a valid CloudLab JWT token.
  • ⚠️The CloudLab JWT token must be saved to a specific file path or provided via the `CLOUDLAB_TOKEN_PATH` environment variable for the server to authenticate.
Verified SafeView Analysis
The server handles CloudLab JWT tokens securely by prioritizing environment variables and local file paths, with clear warnings against committing them to version control. It communicates with the CloudLab API exclusively over HTTPS. No 'eval', code obfuscation, or directly malicious patterns were identified in the provided source code. The server operates via standard I/O (stdio), meaning it does not open its own network ports directly, relying on the invoking application (e.g., Claude Desktop) for communication security. Minor risk: CloudLab API error messages are returned directly and could potentially contain detailed backend information, though this is dependent on the CloudLab API itself.
Updated: 2025-12-06GitHub→
0
0
Medium Cost
kiki830621 icon

che-word-mcp

by kiki830621

Sec8

Programmatically creates, reads, modifies, and exports Microsoft Word (.docx) documents without requiring Microsoft Word installation, providing a wide array of document manipulation tools.

Setup Requirements

  • ⚠️Requires macOS 13.0+ and Swift 5.9+ for building from source.
  • ⚠️Installation involves manual placement of the binary and configuration (e.g., in claude_desktop_config.json or via `claude mcp add`).
Verified SafeView Analysis
The server operates on the local file system using paths provided in tool arguments, which is an inherent risk for any file manipulation tool. While using standard Swift file operations, comprehensive input validation for malicious path traversal (e.g., '..') is crucial and relies on the calling environment or careful agent prompting. The server communicates via standard I/O, limiting direct network exposure. No 'eval' or similar arbitrary code execution patterns were found, and tool dispatch is explicitly handled via a switch statement.
Updated: 2026-01-19GitHub→
0
0
Low Cost
Nami7723 icon
Sec7

This project demonstrates building ChatGPT applications using the Model Context Protocol (MCP) with Next.js, providing tools and resources renderable natively within ChatGPT.

Setup Requirements

  • ⚠️Requires Node.js (>=18.18.0) and pnpm (or equivalent package manager like npm/yarn) for development and deployment.
  • ⚠️Deployment to Vercel is recommended for automatic `baseURL` configuration; otherwise, manual setup of VERCEL_PROJECT_PRODUCTION_URL, VERCEL_BRANCH_URL, or VERCEL_URL environment variables is required.
  • ⚠️OpenAI API Key for ChatGPT interaction is configured within the ChatGPT settings, not directly in this application's environment variables.
Verified SafeView Analysis
The server uses `Access-Control-Allow-Origin: *` in its middleware, which provides broad access for cross-origin requests. While this might be intended for integration with the ChatGPT host, it is generally a security risk that could be tightened for production if specific origins are known. Dynamic client-side JavaScript is injected via `app/layout.tsx` to handle OpenAI SDK integration, but the script is inline and originates from the codebase, not an external untrusted source. No direct `eval` of untrusted user input or hardcoded sensitive secrets found. The base URL is derived from environment variables, which is good practice.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
Koneisto icon
Sec9

Controls Home Assistant lights with detailed color and state management, and creates/manages scenes, complementing the official Home Assistant MCP with advanced light features.

Setup Requirements

  • ⚠️Requires a running Home Assistant instance.
  • ⚠️Requires a Home Assistant long-lived access token.
  • ⚠️Requires Node.js and npm to run the server.
Verified SafeView Analysis
The server design emphasizes safety: it strictly controls only lights, not critical systems. It communicates via stdio, limiting network exposure. Critical light-changing operations require explicit 'user_confirmed=true'. Home Assistant credentials are loaded from environment variables or a local config file, not hardcoded. Local scene backups are maintained, and multi-instance changes are handled with a merge strategy. No 'eval' or malicious patterns were identified.
Updated: 2025-12-21GitHub→
0
0
Medium Cost
auxia-inc icon

mcp-servers

by auxia-inc

Sec7

Manage Google Drive files, folders, Google Sheets, and monitor Drive activity directly from Claude Code.

Setup Requirements

  • ⚠️Requires a graphical environment to open a browser for initial Google OAuth authentication.
  • ⚠️Node.js version >=18 is required to run the server.
  • ⚠️Hardcoded Google OAuth client credentials may need to be overridden if Google revokes or flags them for public exposure. (Though the README indicates these are Auxia's public credentials, users may want to configure their own).
Verified SafeView Analysis
The server uses hardcoded OAuth client ID and client secret in `src/credentials.ts` for Google API access. While the README suggests this is for application identification, hardcoding client secrets in distributed source code is generally discouraged as it can be extracted and potentially misused, even for client-side OAuth flows. However, for local/personal use, the risk is mitigated by user consent. User tokens are stored securely in `~/Claude/.security/gdrive-token.json` with restricted permissions (0o600). Authentication uses a local callback server on `localhost:3035`, which is a standard and relatively secure practice for OAuth flows (PKCE). No 'eval' or obvious malicious patterns were found.
Updated: 2025-12-19GitHub→
0
0
Medium Cost
andytango icon

puppeteer-mcp

by andytango

Sec4

Provides browser automation capabilities through Puppeteer for AI agents to interact with web pages, capture content, and perform various browser operations.

Setup Requirements

  • ⚠️Requires Node.js version 18 or higher to run.
  • ⚠️Puppeteer implicitly downloads a Chromium browser binary on first run, which requires sufficient disk space and network access.
  • ⚠️The default browser launch configuration uses `--no-sandbox`, which is a significant security risk if the server is not run within an isolated container or trusted environment.
Review RequiredView Analysis
The `evaluate` tool allows arbitrary JavaScript code execution via `new Function()`. If a malicious or compromised AI agent injects code, it could execute within the browser context. More critically, Puppeteer is launched with `--no-sandbox` which significantly reduces browser security. This makes the server vulnerable to host system compromise if not run within a highly isolated environment (e.g., a secure Docker container) as suggested in the README for remote deployments. No obvious hardcoded secrets were found in the truncated source.
Updated: 2025-12-03GitHub→
0
0
Low Cost
rao-1706 icon
Sec7

An MCP server providing analytics tools, secured by Keycloak-based OAuth2 authentication.

Setup Requirements

  • ⚠️Requires a running Keycloak server, configured with a specific realm, client, and custom scopes (`mcp:tools`, `mcp:resources`), as detailed in `docker/keycloak/KEYCLOAK_SETUP_GUIDE.md`.
  • ⚠️Python 3.12 or newer is required.
  • ⚠️Authentication is disabled by default (`ENABLE_AUTH=false`); setting `OAUTH_CLIENT_SECRET` and `ENABLE_AUTH=true` is necessary for active authentication.
Verified SafeView Analysis
The `verify_token` function explicitly skips scope and audience validation ('no scope/audience validation needed for dev'), making authorization less robust for production use cases. Keycloak's `docker-compose.yml` uses default, insecure credentials ('admin'/'admin'), which should never be used in production. The system relies on correct configuration of `OAUTH_CLIENT_SECRET` via environment variables.
Updated: 2025-12-02GitHub→
0
0
Medium Cost
Amardeshmukh2566 icon

test-remote-mcp-server

by Amardeshmukh2566

Sec1

A server for testing remote interactions, possibly related to a Multiple-Choice Protocol (MCP).

Review RequiredView Analysis
No source code was provided for analysis. Therefore, a comprehensive security audit for 'eval', obfuscation, network risks, hardcoded secrets, or malicious patterns could not be performed. The score reflects the inability to assess safety.
Updated: 2026-01-17GitHub→
0
0
Medium Cost
akraradets icon

boilerplate-mcp

by akraradets

Sec3

This repository serves as a boilerplate/template for developing a Model Context Protocol (MCP) server, integrating tools, resources, and prompts for LLM applications.

Setup Requirements

  • ⚠️Requires Python 3.13 or newer.
  • ⚠️FastMCP is a stateful framework, which can impact horizontal scaling and might lead to loss of elicitation/sampling features when behind a load balancer without sticky sessions.
  • ⚠️Requires Node.js and npm/npx for `MCP Inspector`.
Review RequiredView Analysis
The server exposes a resource template `files://{story_name}` which directly uses the `story_name` from client requests to construct file paths. Without explicit sanitization or validation, this is vulnerable to path traversal (e.g., `files://../../../../etc/passwd`), potentially allowing unauthorized reading of arbitrary files on the server. The `allow_origins=["*"]` in CORS settings is also overly permissive for production.
Updated: 2025-12-27GitHub→
0
0
Medium Cost
trashchenkov icon

nasa_mcp

by trashchenkov

Sec9

The server provides a collection of tools to interact with various NASA APIs for astronomy pictures, space weather, media search, and near-earth objects, acting as a unified interface for agent-based systems.

Setup Requirements

  • ⚠️Requires 'fastmcp' Python library to be installed.
  • ⚠️Requires 'NASA_API_KEY' environment variable for full functionality, otherwise a rate-limited 'DEMO_KEY' is used.
Verified SafeView Analysis
The server retrieves the NASA API key from an environment variable, using a 'DEMO_KEY' as a fallback, which is a good practice. It utilizes `httpx` with `raise_for_status()` for robust HTTP requests. No direct `eval` or other arbitrary code execution patterns are visible. Input parameters are mostly handled by the underlying NASA APIs, and string inputs for `event_type` in `nasa_donki_recent_events` are passed directly after uppercasing, which is generally safe within the context of predefined NASA API event types.
Updated: 2025-12-22GitHub→
0
0
Medium Cost
nickweedon icon
Sec7

Provides AI assistants with tools to read, create, edit, and manage Google Docs and Google Drive files, including advanced formatting and bulk operations.

Setup Requirements

  • ⚠️Requires a Google Cloud Project setup, including enabling Google Docs and Drive APIs, configuring an OAuth Consent Screen, and creating OAuth Desktop app credentials (a multi-step manual process).
  • ⚠️The Docker socket (`/var/run/docker.sock`) must be mounted into the container (read-only) for the automatic OAuth port discovery feature to function.
  • ⚠️A blank `credentials/token.json` file needs to be created manually before the initial Google authentication process.
Verified SafeView Analysis
The server requires mounting the Docker socket (`/var/run/docker.sock`) into the container as read-only. While this mitigates some risks, access to the Docker socket can still be a significant security concern if the container is compromised, potentially allowing privilege escalation on the host system. The application requests broad Google Drive API permissions (`https://www.googleapis.com/auth/drive`), granting extensive access to a user's Drive content. Users should carefully review the implications of these permissions and ensure they are comfortable with the scope. No hardcoded secrets, `eval()` calls, or obvious malicious patterns were found in the provided source code. Input validation and error handling are present to prevent common API abuse and provide user-friendly messages.
Updated: 2025-12-25GitHub→
0
0
Medium Cost

memory-mcp-server

by win10ogod

Sec7

A server component designed for managing and processing an AI system's memory, including features for image optimization and handling prompts for language models.

Setup Requirements

  • ⚠️Requires Node.js dependencies to be installed (`npm install`)
  • ⚠️Requires configuration based on `example-config.json`
  • ⚠️Likely requires API keys for external AI/LLM/Vision services (e.g., OpenAI, Google Cloud AI) for full functionality
Verified SafeView Analysis
Cannot perform full code analysis for 'eval' or obfuscation. Assumes standard Node.js security practices based on file structure. Network risks are inherent to any server application, but no obvious malicious files or patterns were detected from the filenames.
Updated: 2025-11-18GitHub→
PreviousPage 590 of 760Next