Stop Searching. Start Trusting.

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

Browse Directory Why 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)

Low Cost

AI-Infra

by Bruno1702vm

Sec2

Provides an application for exploring and learning about AI infrastructure, particularly LLM inference engines and Kubernetes concepts.

Setup Requirements

⚠️Requires Windows 10+ / macOS Catalina+ / Linux (Ubuntu 18.04+)
⚠️Requires at least 4 GB RAM and 500 MB free disk space
⚠️Requires downloading and installing an unverified binary from a raw GitHub URL.

Review RequiredView Analysis

The provided source code largely consists of markdown documentation. However, the `README.md` instructs users to download and run a pre-compiled application (`AI-Infra.zip`) directly from a raw GitHub URL. The source code for this specific application is not provided for analysis. Running binaries from unverified sources without access to their source code for auditing is a critical security vulnerability and poses substantial risks, including the potential for malware, system compromise, or data exfiltration. Therefore, from a security standpoint, the project, as presented with a downloadable binary, is highly unsafe to execute without further inspection of the binary itself or its source code.

Updated: 2026-01-19GitHub→

Medium Cost

alfa-pro-terminal-mcp

by Fredoq

Sec9

A Model Context Protocol server that exposes the Alfa Investments PRO Terminal API to LLM clients over stdio.

Setup Requirements

⚠️Requires the Alfa Investments PRO Terminal desktop application to be running locally.
⚠️Requires .NET SDK (version 10.0.100 or compatible with C# 14, as specified in global.json and AGENTS.md - note that .NET 10/C# 14 are future versions, indicating a forward-looking or custom environment).
⚠️The 'dnx' command used in the quick start might be a custom alias or specific runtime environment, as 'dnx' was an older .NET Core runtime.

Verified SafeView Analysis

The server connects to a local WebSocket endpoint (ws://127.0.0.1:3366/router/) of the Alfa Investments PRO Terminal desktop app, minimizing external network attack surface. Configuration can be overridden by environment variables (TERMINAL__ENDPOINT, TERMINAL__TIMEOUT). No obvious hardcoded secrets, obfuscation, or 'eval' usage in the provided snippets. The project includes a SECURITY.md file detailing vulnerability reporting. While appsettings.json has 'AllowedHosts: "*"', this is less critical for a local stdio-based MCP server whose primary interaction is not through HTTP. The AGENTS.md document also describes good security-conscious coding practices (e.g., 'fail fast', no reflection for business logic).

Updated: 2026-01-13GitHub→

Low Cost

Agentic-Emotion

by sean-jack

Sec9

An MCP server that enables Claude Code to analyze text and display emotions in a real-time visual window.

Setup Requirements

⚠️Requires .NET 8 SDK or higher for the Avalonia display application to run.
⚠️Requires Python 3.9+ and can involve potentially large `transformers` and `torch` model downloads on the first run of the MCP server.
⚠️Configuration involves manually editing `claude_desktop_config.json` with absolute paths, which is platform-specific (e.g., forward vs. backslashes, `python` vs `python3` executable name).
⚠️The Avalonia display application must be running separately (in a different terminal or as a background process) for emotions to be visually displayed.

Verified SafeView Analysis

The server uses local inter-process communication (named pipes on Windows, Unix sockets on macOS/Linux) and `stdio` for MCP communication, significantly limiting network exposure. Text input for emotion analysis is processed by a local AI model (`transformers` pipeline) and keyword matching, without direct `eval` or arbitrary code execution. `pywin32` is a dependency on Windows, which interacts with OS internals but is a standard library.

Updated: 2025-12-03GitHub→

Low Cost

confirminator

by fairanswers

Sec9

An MCP Server offering tools to confirm the successful compilation and testing of Java/Maven projects.

Setup Requirements

⚠️Requires Java Development Kit (JDK) to build and run.
⚠️Requires Maven for local development and build.
⚠️Designed to be part of an MCP (Multi-Agent Communication Protocol) ecosystem; its full utility might require an MCP runtime.

Verified SafeView Analysis

The current implementation of `ConfirmJavaMavenTool.confirm()` is extremely minimal, logging a message and returning a hardcoded string, thus posing low security risk. The commented-out `@McpToolParam File zipFile` suggests future functionality involving file processing, which would require careful security considerations (e.g., zip bomb, path traversal, arbitrary code execution through build scripts) if implemented.

Updated: 2025-11-18GitHub→

Low Cost

IA_II-MCP_Server

by gabrielbergel

Sec9

This project simulates a Multi-Agent Coordination Platform (MCP) agent for automated IT support, demonstrating how an agent processes user intentions, applies guardrails, executes specialized tools (diagnostics, resolution, escalation), and logs all operations.

Verified SafeView Analysis

The code does not contain 'eval', obfuscation, or hardcoded secrets. It implements input and output guardrails for the 'escalate_ticket' function, which is a good practice. All operations are logged to a local JSON file ('operation_log.json'). While the logging itself is secure (local file, no network exposure), the content of the logs could potentially contain sensitive user information if not properly managed in a production environment. However, for a simulated environment, this is acceptable. There are no direct network risks.

Updated: 2025-11-20GitHub→

Medium Cost

mcp-rag-vector

by AppliNH

Sec9

An MCP server designed to serve as a RAG pipeline, enabling LLMs to write and read from a vector database.

Setup Requirements

⚠️Requires local Ollama server running
⚠️Requires Docker and Docker Compose
⚠️Requires downloading large LLM models (e.g., 2 GB)

Verified SafeView Analysis

The system binds HTTP servers to `0.0.0.0`, which is a common practice within Docker containers but requires careful network configuration in production environments. The 'greet' tool processes user input by simple string concatenation, which is safe. No 'eval', direct shell execution of unsanitized user input, or obvious hardcoded secrets were identified in the provided code snippet.

Updated: 2025-12-05GitHub→

Low Cost

onenotemcp

by IAmKSquared

Sec9

The OneNote MCP Server enables AI assistants to securely read, write, search, and comprehensively edit Microsoft OneNote data via the Model Context Protocol.

Setup Requirements

⚠️Requires an active Microsoft account with OneNote access.
⚠️For production/shared use, it is strongly recommended to create your own Azure Application Registration and set the AZURE_CLIENT_ID environment variable with specific delegated permissions.
⚠️Initial setup requires an interactive browser authentication via the Microsoft device code flow.
⚠️When connecting to an MCP client (e.g., Claude Desktop), the absolute path to `src/server.mjs` must be specified correctly.

Verified SafeView Analysis

The project demonstrates strong security practices: it uses OS-native secure storage (keyring) for encryption keys, encrypts access tokens using AES-256-GCM, employs Zod for robust input validation and sanitization to prevent injection attacks, and explicitly disables `no-new-func` in ESLint to prevent dangerous code execution. While it defaults to a public Azure Client ID for testing, it strongly recommends using a custom, securely configured Azure App Registration via environment variables for production use. No `eval` or similar dangerous patterns were found in the provided code snippets.

Updated: 2025-12-27GitHub→

Medium Cost

mail-mcp

by martinzarfl

Sec2

Enables AI assistants to securely send, read, manage, and organize emails via SMTP and IMAP.

Setup Requirements

⚠️Requires Node.js version 23.0.0 or higher.
⚠️Requires configuring multiple environment variables for SMTP and IMAP connections (e.g., host, port, user, password).
⚠️The IMAP connection's hardcoded `rejectUnauthorized: false` disables TLS certificate validation, posing a significant security risk for MITM attacks. This must be addressed for secure use.
⚠️May require generating app-specific passwords for email providers (e.g., Gmail) if 2-Factor Authentication is enabled.

Review RequiredView Analysis

The server hardcodes `tlsOptions: { rejectUnauthorized: false }` for IMAP connections within `createImapConnection`. This critical vulnerability disables TLS certificate validation, making IMAP connections susceptible to Man-in-the-Middle (MITM) attacks, even if TLS is ostensibly enabled. While credentials are managed via environment variables and the SSE transport supports OAuth 2.0, this IMAP flaw severely compromises the overall security for sensitive email data.

Updated: 2025-11-19GitHub→

Medium Cost

mcp-server-tasktracking

by Devanand-debug-ai

Sec7

Manages tasks using an Excel spreadsheet as the backend, exposing CRUD operations and statistics via an MCP server with a conversational interface.

Setup Requirements

⚠️Requires a Python 3.x environment with 'mcp' and 'openpyxl' libraries installed.
⚠️Needs write permissions in the script's directory for 'tasks.xlsx' to be created or modified.
⚠️Designed to be run by an MCP-compatible client (e.g., Claude Desktop) as it uses standard I/O for communication.

Verified SafeView Analysis

The server reads and writes to a local 'tasks.xlsx' file based on user tool calls. While input is validated via 'inputSchema' to prevent arbitrary tool execution, malicious Excel formulas or excessively long strings could potentially be injected into cells (e.g., task title, category). This could lead to data corruption or unintended behavior if the Excel file is later opened by another application without proper sanitization. However, there is no direct arbitrary code execution vulnerability identified within the Python server process itself through user input.

Updated: 2025-11-24GitHub→

Low Cost

SimplyBook-Mcp-Server

by Umair-Siddique

Sec8

The SimplyBook-Mcp-Server acts as a FastMCP server to provide a structured interface for interacting with the SimplyBook.me REST API, facilitating booking and service management.

Setup Requirements

⚠️Requires Docker Desktop for the recommended deployment method using Docker Compose.
⚠️SimplyBook.me credentials (`SIMPLYBOOK_COMPANY`, `SIMPLYBOOK_LOGIN`, `SIMPLYBOOK_PASSWORD`) are mandatory environment variables.
⚠️Ngrok must be installed and authenticated if public access is desired, noting free-tier limitations like changing URLs and session timeouts.
⚠️Requires Python 3.11+ for local execution.

Verified SafeView Analysis

The server handles SimplyBook.me API credentials via environment variables. The obtained X-Token is stored in a temporary file (cleared after 1 hour), which is a minor risk but mitigated by short expiry. Sensitive data is sanitized in logs before writing. No direct 'eval' or obfuscation is present. Public exposure via Ngrok is possible, and documentation explicitly warns about associated security risks, indicating awareness but placing responsibility on the user for secure deployment.

Updated: 2026-01-16GitHub→

Medium Cost

mcp-server

by dimmark2

Sec2

Provides an AI agent with tools to query and explore a Postgres database schema (specifically 'df365') via the Model Context Protocol (MCP).

Setup Requirements

⚠️Requires a running PostgreSQL database instance.
⚠️Database connection details (host, port, user, password, database) are primarily configured via environment variables, but insecure hardcoded defaults exist.
⚠️Relies on the `@modelcontextprotocol/sdk` for client-server communication.

Review RequiredView Analysis

CRITICAL: The server has a hardcoded default PostgreSQL password ('qDJqEEbhMrQThzXAKRgtIFzFVKsHSaio') which is a severe security vulnerability. CRITICAL: The `sample_rows` and `describe_table` tools are vulnerable to SQL injection as they construct SQL queries by directly concatenating user-provided `schema` and `table` names without proper sanitization or parameterization for identifiers. The `run_select` tool attempts to prevent data modification and arbitrary SQL execution via a blacklist, but this approach is less secure than whitelisting or using parameterized queries for dynamic identifiers, and does not cover the other vulnerable tools.

Updated: 2025-12-06GitHub→

Medium Cost

kickflow-mcp-server

by kickflow

Sec9

This MCP server allows AI agents to interact with the kickflow API for workflow automation and data management.

Setup Requirements

⚠️Requires Node.js v22.18.0 or higher.
⚠️Requires a Kickflow Access Token (KICKFLOW_ACCESS_TOKEN) to be set as an environment variable or passed as a command-line argument.

Verified SafeView Analysis

The server retrieves the Kickflow access token from environment variables or command-line arguments, avoiding hardcoding of sensitive credentials. It uses Zod for input validation on API parameters, enhancing robustness against malformed requests. API interactions are handled via a well-established Axios library. File uploads from base64 strings are processed carefully, minimizing direct execution risks.

Updated: 2026-01-19GitHub→

PreviousPage 391 of 760Next