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)

30
1
Low Cost
jtwolfe icon
Sec8

Provides AI models and applications with structured access to OParl parliamentary data APIs via the Model Context Protocol.

Setup Requirements

  • ⚠️Requires Python 3.11 or higher.
  • ⚠️Requires 'oparl_openapi.json' file to be present in the project root for server initialization.
  • ⚠️Relies on an external OParl API endpoint (defined by OPARL_BASE_URL); connectivity and API key validity are crucial for functionality.
Verified SafeView Analysis
The server uses `pydantic_settings` for configuration, allowing API keys to be loaded securely from environment variables or `.env` files, rather than being hardcoded. It integrates with `httpx` for making asynchronous API calls to external OParl APIs, with timeouts configured. There is no usage of `eval` or similar dangerous functions that could lead to immediate remote code execution. The system relies on an `oparl_openapi.json` file for its API definition; ensuring this file is trusted is important. Outbound network requests are fundamental to its operation, so the security of the target OParl API is an external dependency. Overall, the codebase appears to follow good security practices for its architectural pattern.
Updated: 2025-12-15GitHub→
30
1
Medium Cost
omise icon

omise-mcp

by omise

Sec9

Facilitate secure and managed integration with Omise payment APIs for various financial operations using the Model Context Protocol.

Setup Requirements

  • ⚠️Requires Node.js 20+ to run directly or Docker/Podman for containerized deployment.
  • ⚠️Mandatory Omise Account and API Keys (`OMISE_SECRET_KEY`, `OMISE_ENVIRONMENT`) must be configured.
  • ⚠️Mandatory tool access control (`TOOLS` environment variable) must be explicitly set ('all' or a comma-separated list of tool names). The server will not start without this.
Verified SafeView Analysis
The server demonstrates strong security practices including explicit tool-level access control via the 'TOOLS' environment variable, which is mandatory for startup. It validates all incoming parameters for API calls, significantly reducing attack surface. Sensitive data like API keys, passwords, and card numbers are actively masked in logs using `sanitizeHeaders` and `sanitizeBody` functions. The Docker configuration enforces hardened practices like running as a non-root user, using `no-new-privileges`, `read_only` file systems, and `tmpfs` for temporary directories. The `OMISE_SECRET_KEY` is externalized and validated for environment consistency (e.g., preventing live keys in test mode). Comprehensive error handling and logging (with redaction) are in place. No 'eval' or obvious obfuscation found.
Updated: 2025-11-29GitHub→
30
1
High Cost
codeiva4u icon
Sec6

A powerful multi-protocol (MCP, LSP, SSE) server that enables AI agents and other clients to automate Brave Real Browser, featuring anti-detection, ad-blocking, stealth, and advanced web interaction capabilities for complex scraping, data extraction, and workflow automation.

Setup Requirements

  • ⚠️Requires Brave Browser to be installed on the host machine (though brave-real-launcher attempts auto-installation).
  • ⚠️Requires Node.js version 18.0.0 or higher.
  • ⚠️The `execute_js` tool grants significant control over the browser, requiring strict trust in the input from the calling AI agent or client to prevent malicious code execution.
Verified SafeView Analysis
The server includes an `execute_js` tool, which is an 'ULTRA POWERFUL' feature allowing custom JavaScript code execution directly on the browsed page. This grants high privileges within the browser context and poses a significant security risk if the input to this tool is not from a trusted source or is vulnerable to injection attacks. While `execSync` is used in some system utilities and browser launching components, it is generally for controlled system-level operations like process management and file operations with explicit path sanitization in `file-handlers.ts`. No direct `eval` calls of external input were found in the main processing logic. However, the presence of `execute_js` makes the system highly capable but also highly sensitive to the trustworthiness of the calling agent.
Updated: 2026-01-19GitHub→
30
1
Medium Cost
bajoski34 icon

mcp-flutterwave

by bajoski34

Sec9

Enables AI assistants to interact with Flutterwave for payment processing, transaction management, and financial operations.

Setup Requirements

  • ⚠️Requires a Flutterwave Secret Key (FLW_SECRET_KEY) obtained from the Flutterwave dashboard.
  • ⚠️Requires Node.js v18 or higher.
  • ⚠️The 'mcp-flutterwave' command line tool requires the '--tools' argument specifying which tools to enable (e.g., '--tools=all').
Verified SafeView Analysis
The server correctly utilizes environment variables (FLW_SECRET_KEY) for API authentication, avoiding hardcoded credentials. It leverages openapi-fetch for API interactions, which provides type safety and a structured approach to external calls. Input validation is performed using Zod schemas for tool arguments. No eval or similar dangerous functions are observed without clear justification.
Updated: 2026-01-16GitHub→
30
1
High Cost
dabster108 icon
Sec8

An AI-powered recommendation system for students, providing personalized content (forums, courses, events, scholarships, etc.) and integrating with an AI assistant via the Model Context Protocol (MCP).

Setup Requirements

  • ⚠️Requires Python 3.11+
  • ⚠️Requires Google Gemini API Key (paid, for AI assistant and semantic checks)
  • ⚠️Requires Groq API Key (paid, for enhanced keyword extraction) for full functionality, though the current `app/main.py` uses Gemini for keyword extraction.
  • ⚠️The FastAPI server must be running before the MCP server and client can successfully start and connect to it.
Verified SafeView Analysis
The project correctly uses environment variables for API keys (GEMINI_API_KEY, GROQ_API_KEY) and implements robust Pydantic-based data validation, significantly reducing common web vulnerabilities. However, the CORS `allow_origins=["*"]` is a security risk in production environments and should be restricted to known origins. Student data is stored in flat JSON files (`student.json`), which is not suitable for production and lacks proper access control or encryption. Interaction with external LLM APIs (Google Gemini) introduces dependency on third-party services for data handling (user queries, content items). No direct `eval` or blatant malicious patterns were found.
Updated: 2025-12-04GitHub→
30
1
Medium Cost
recallium-ai icon

recallium

by recallium-ai

Sec9

Recallium acts as a persistent memory layer for AI agents in IDEs, providing cross-project intelligence, document knowledge base integration, and structured memory recall to enhance AI coding assistants.

Setup Requirements

  • ⚠️Requires Docker to run the server and PostgreSQL database.
  • ⚠️Requires Ollama to be installed and running locally, or API keys for OpenAI/Anthropic for LLM capabilities.
  • ⚠️For command-only IDEs (e.g., Zed, JetBrains, Claude Desktop), the `@recallium/mcp-client` npm package is required.
Verified SafeView Analysis
The architecture is privacy-first, designed to run locally, keeping data on the user's machine. It uses Docker for isolation. The `POSTGRES_PASSWORD` has a default value (`recallium_password`) in `recallium.env`, but the documentation explicitly warns to 'Change in production!'. No `eval` or blatant malicious patterns are evident in the provided source code (READMEs).
Updated: 2026-01-18GitHub→
30
1
High Cost
pasie15 icon
Sec7

Provides an MCP interface to Scenario.com's generative AI API for tasks like text-to-image, image-to-image generation, model training, asset management, and image processing.

Setup Requirements

  • ⚠️Requires Scenario.com API Key and Secret (paid service) for authentication.
  • ⚠️When not using environment variables, the 'set-credentials' tool must be explicitly called first in each session to configure API access.
  • ⚠️Node.js runtime environment is required.
Verified SafeView Analysis
The server acts as a proxy to the Scenario.com API. Authentication relies on API keys provided via environment variables or session-based tools, which are handled in-memory and not persisted to disk. URL path and query parameters are constructed from tool arguments, relying on axios for proper encoding and the upstream API for input validation. The 'post-generate-custom' tool allows passing an arbitrary body, which could be a risk if the backend API is vulnerable to malicious JSON payloads. No direct use of 'eval' or unprotected shell command execution found.
Updated: 2025-11-23GitHub→
30
66
Medium Cost
jlbadano icon

ig-mcp

by jlbadano

Sec8

Integrate AI applications with Instagram Business accounts to manage profiles, media, insights, and direct messages.

Setup Requirements

  • ⚠️Requires an Instagram Business Account connected to a Facebook Page.
  • ⚠️Direct Messaging features (`instagram_manage_messages`) require Advanced Access approval from Meta via a potentially lengthy App Review process (3-7 business days or more).
  • ⚠️Requires generating a long-lived Instagram Access Token, which involves a multi-step process through Facebook Developer tools.
Verified SafeView Analysis
The project follows good security practices: secrets are loaded from environment variables (no hardcoded credentials), input validation is present for tool arguments, `httpx` is used with timeouts and connection limits, and structured logging is implemented. Rate limiting helps prevent abuse. The tool that downloads images for validation (`_validate_image_aspect_ratio`) could be a minor concern if `image_url` can be controlled by a malicious external actor leading to potential SSRF or resource exhaustion, but within the typical MCP context where an AI provides the input, this risk is mitigated.
Updated: 2025-12-22GitHub→
30
1
Medium Cost
Knuckles-Team icon

archivebox-api

by Knuckles-Team

Sec7

Provides an AI-agent-friendly Model Context Protocol (MCP) server and Python wrapper for interacting with an ArchiveBox web archiving instance, enabling programmatic access and automation.

Setup Requirements

  • ⚠️Requires a separate running ArchiveBox instance to connect to.
  • ⚠️Requires an LLM provider (e.g., OpenAI, Anthropic, Google, HuggingFace, or a local Ollama instance) and corresponding API keys, which may incur costs or require local setup.
  • ⚠️Complex authentication and authorization configurations are supported, requiring careful setup to avoid security vulnerabilities.
Verified SafeView Analysis
The project uses `pickle` for model serialization, which is a known vulnerability if untrusted data is deserialized (e.g., if an attacker can control the file path in `load_model`). The `ARCHIVEBOX_VERIFY` environment variable can be set to `False`, disabling SSL certificate validation and exposing the system to Man-in-the-Middle attacks. Default `admin/admin` credentials in `compose.yml` are weak and must be changed. The multiple authentication and authorization mechanisms (JWT, OAuth, OIDC, Eunomia) are powerful but add significant complexity, increasing the risk of misconfiguration.
Updated: 2026-01-19GitHub→
30
1
Low Cost
canstralian icon

parrot_mcp_server

by canstralian

Sec3

Provides a robust Bash script library and a minimal MCP server stub for Raspberry Pi 5, focused on system automation, maintenance, and interaction with AI agents.

Setup Requirements

  • ⚠️The core 'MCP server' is a minimal stub that polls files in `/tmp` by default, not a network-listening daemon. Expect a simulation, not a full server implementation out-of-the-box.
  • ⚠️Requires manual configuration (`PARROT_IPC_DIR`) to move IPC out of `/tmp` for even minimal security, as the default is critically insecure.
  • ⚠️Many scripts rely on `sudo` for system operations, requiring the user to have appropriate `sudo` privileges configured.
  • ⚠️External tools like `jq`, `bc`, and `mail` (mailutils) are required for full functionality and validation features.
Review RequiredView Analysis
The server has critical security vulnerabilities as documented by the project itself: 1. Insecure Inter-Process Communication (IPC) via world-readable files in /tmp, prone to race conditions and symlink attacks. The 'start_mcp_server.sh' still uses this vulnerable default. 2. No Authentication or Authorization for the MCP server, allowing any local process to send commands. 3. Insecure Cron Setup using predictable temporary files. While significant progress has been made on input validation functions in 'common_config.sh', not all scripts fully utilize them or inherent risks remain due to the fundamental architecture of the server stub.
Updated: 2026-01-09GitHub→
30
1
Low Cost
demiliani icon
Sec8

A Visual Studio Code extension to automate the installation and configuration of the Dynamics 365 Business Central Admin MCP server for GitHub Copilot integration.

Setup Requirements

  • ⚠️Requires .NET 9 runtime to be installed for the actual MCP server (separate from Node.js/npm).
  • ⚠️Global npm package installation may require elevated permissions (admin/sudo) depending on system configuration.
Verified SafeView Analysis
The extension uses `child_process.exec` and `execSync` to run system commands (npm install/update/uninstall, Node.js/npm version checks, process management). While commands are mostly hardcoded or derived from trusted sources, installing global npm packages introduces a supply chain risk (dependency on `@demiliani/d365bc-admin-mcp`). File system operations are confined to expected configuration (`mcp.json`) and temporary directories. No direct use of `eval` or obvious obfuscation. Process termination logic (`taskkill`, `pkill`) targets specific executables, reducing broader system risk. The `.NET 9` requirement for the *installed* MCP server (not this extension itself) implies a dependency on another runtime. Overall, the implementation appears robust for its stated purpose, but inherent risks of running system commands and external package installations exist.
Updated: 2025-12-11GitHub→
30
1
Medium Cost
aplaceforallmystuff icon

mcp-pihole

by aplaceforallmystuff

Sec9

Connects AI assistants to Pi-hole to manage DNS blocking, view statistics, and control ad-blocking features via natural language.

Setup Requirements

  • ⚠️Requires Pi-hole v6 with its API enabled.
  • ⚠️Requires a Pi-hole app password, which must be manually generated from the Pi-hole web interface.
  • ⚠️Requires Node.js 18+ to run.
  • ⚠️Requires 'PIHOLE_URL' and 'PIHOLE_PASSWORD' environment variables to be explicitly set.
Verified SafeView Analysis
The server primarily acts as a wrapper around the Pi-hole v6 API. It correctly handles the Pi-hole password by requiring it as an environment variable (PIHOLE_PASSWORD) rather than hardcoding it. No 'eval' or direct arbitrary command execution via 'child_process' is observed. Network requests are made using 'fetch' to the user-configured PIHOLE_URL. The primary security consideration relies on the user providing a trusted PIHOLE_URL, as a malicious configuration could redirect the Pi-hole password to an untrusted endpoint. However, this is a configuration risk, not a code vulnerability within the server itself. Input validation for tools relies on the MCP SDK's schema definitions (e.g., 'type: string') and the Pi-hole API's internal handling of domain names, which is standard practice.
Updated: 2026-01-16GitHub→
PreviousPage 215 of 760Next