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)

Medium Cost

gemini-image-mcp

by lordmoocow

Sec4

This server provides an MCP (Multi-Component Platform) interface for generating images using Google's Gemini 3 Pro Image model.

Setup Requirements

⚠️Requires Google API Key (GOOGLE_API_KEY environment variable) for Google Gemini API, which is a paid service.
⚠️Requires Python 3.10 or newer.
⚠️Designed to run within an MCP framework, communicating via standard I/O.

Review RequiredView Analysis

The `call_tool` function allows users to specify an arbitrary `file_path` for saving generated images. This path is used directly with `pathlib.Path(file_path)` and `path.write_bytes()`. This presents a significant directory traversal vulnerability, as a malicious user could specify paths like `../../../../tmp/malicious.png` to write files to arbitrary locations on the server's filesystem, potentially overwriting critical files or planting malicious executables, especially if the server runs with elevated permissions. The `path.resolve()` call occurs after the file write and only resolves the path for reporting, not for preventing the write itself. No input sanitization for `file_path` is evident.

Updated: 2025-11-25GitHub→

Medium Cost

weather-mcp-a2a

by manuelalba1021

Sec8

An agentic AI system that fetches and reasons over real-time global weather data using the Model Context Protocol (MCP) and large language models.

Setup Requirements

⚠️Requires GROQ_API_KEY for the LLM inference (paid service).
⚠️Requires Python 3.11 or later.
⚠️Requires active internet connection for real-time data fetching.
⚠️Microphone permissions needed for voice input functionality.

Verified SafeView Analysis

The project adheres to good security practices by loading API keys from environment variables (`.env`). External API calls are made using `httpx` and `requests` to well-known weather APIs (`api.weather.gov`, `open-meteo.com`), which mitigates direct network risks. Input sanitization for city names used in API requests relies on URL parameter encoding by the `requests` library, which is generally robust against injection. No use of `eval` or other dangerous code execution patterns was identified. The most significant concern, noted in the project's internal documentation, is the unreliability of certain LLMs (Groq models) in correctly formatting tool calls, which led to a direct API implementation workaround in the main Streamlit app. This workaround itself is implemented securely, but deviates from the intended agentic workflow.

Updated: 2026-01-19GitHub→

Low Cost

MultiAgent-with-A2A-MCP-server

by CaptainMaxine

Sec6

This server provides a communication backbone for a multi-agent system, facilitating both external client-to-server and internal agent-to-agent communication via WebSockets and MQTT.

Setup Requirements

⚠️Requires Python 3+ environment and `pip install -r requirements.txt`
⚠️Likely requires a running MQTT broker (e.g., Mosquitto) for agent-to-agent communication, as indicated by `asyncio-mqtt`.

Verified SafeView Analysis

No explicit 'eval', 'exec', or hardcoded secrets were found in the truncated source code. The server binds to '127.0.0.1' (localhost) by default, which is safer for local testing. However, as a network server, inherent risks exist without clear authentication/authorization mechanisms and robust input validation, especially within 'message_handling.py'. The use of MQTT for agent-to-agent communication also implies the need for a secured MQTT broker to prevent unauthorized agent impersonation.

Updated: 2025-11-30GitHub→

Low Cost

awesome-mcp

by Rodert

Sec8

This project is an automated generator and host for a curated list of Model Context Protocol (MCP) projects, providing an interactive and searchable directory, rather than being an MCP server itself.

Setup Requirements

⚠️Requires Python 3.11+ and Node.js 18+.
⚠️Requires a GitHub Personal Access Token (`GITHUB_TOKEN`) with `public_repo` permissions configured as an environment variable for project scraping.
⚠️Initial run of the translation script (`translate_content.py`) will download large language packages for offline translation, which can take time and consume bandwidth.

Verified SafeView Analysis

The project functions as an automated 'awesome list' generator for MCP projects and a VitePress static site. Its Python scripts interact with the GitHub API, requiring a `GITHUB_TOKEN` which is correctly managed via an environment variable. No hardcoded secrets or 'eval' statements were found. The `translate_content.py` script uses an offline translation library (`argostranslate`), minimizing external API calls for translation. The main security considerations are external to this repository's codebase and relate to how users configure *other* MCP servers (e.g., `GITHUB_PERSONAL_ACCESS_TOKEN` for `modelcontextprotocol/server-github`) as described in the README's quick-start guides.

Updated: 2026-01-19GitHub→

Low Cost

arcade-mcp-ts

by ArcadeAI

Sec8

A framework for building servers that provide tools for AI assistants, enabling them to interact with external services and data.

Setup Requirements

⚠️Node.js 20+ runtime environment required for the TypeScript framework, utilizing ESM (ECMAScript Modules) only.
⚠️Requires external 'arcade-mcp' CLI for generating new Python server projects and for authentication setup (e.g., 'arcade login').
⚠️Specific tools built with the framework will require configuration of secrets (e.g., MY_SECRET_KEY) as environment variables or OAuth tokens, as demonstrated in the Python examples.
⚠️This repository (arcade-mcp-ts) is a server framework/library, not a standalone executable server, meaning there isn't a single 'run' command for the framework itself; rather, developers use it to build their own servers. The provided README.md primarily details how to run a *generated Python server* using the 'arcade-mcp' CLI.

Verified SafeView Analysis

The framework emphasizes secure handling of secrets and OAuth tokens, explicitly stating they are injected into the context at runtime and are not visible to LLMs or MCP clients. This design pattern reduces the risk of sensitive data exposure. No explicit malicious patterns like 'eval' or obfuscation were found in the provided truncated source code. A dedicated SECURITY.md also points to a security research program.

Updated: 2025-12-02GitHub→

Low Cost

flux-mcp

by tehw0lf

Sec8

A server and CLI tool for generating high-quality images using FLUX models, featuring VRAM management and automatic model unloading.

Setup Requirements

⚠️Requires an NVIDIA GPU with 12GB+ VRAM (16GB+ recommended) and CUDA toolkit installed.
⚠️Requires PyTorch with CUDA support, which can be complex to set up correctly.
⚠️Configuration relies on a `.env` file (e.g., `FLUX_OUTPUT_DIR`) which must be manually copied and edited.

Verified SafeView Analysis

The code appears robust in handling user input, with parameters like steps, guidance, and dimensions being type-cast and range-validated before use in the generation pipeline. File paths for output are derived from configuration or validated CLI arguments, reducing path traversal risks. System calls (`subprocess.run` for opening file managers) use controlled arguments (a `Path` object), mitigating command injection. There are no obvious hardcoded secrets or direct `eval()`/`exec()` calls on user-controlled input. The primary potential vectors for vulnerability would lie in the underlying `diffusers` library or malicious model weights from HuggingFace, which are outside the scope of this code's direct security implementation.

Updated: 2025-12-07GitHub→

Low Cost

mcp_servers

by SvenTern

Sec7

Synchronizes MCP server configurations from a YAML file to the Claude CLI and generates a project-level .mcp.json configuration.

Setup Requirements

⚠️Requires 'claude' CLI to be installed and accessible in PATH.
⚠️Requires a `mcp-servers.yaml` file in the current directory or script directory for configuration.
⚠️Authentication tokens for MCP servers will be read from the `mcp-servers.yaml` file, potentially storing sensitive information in plaintext.

Verified SafeView Analysis

The script itself does not contain 'eval' or malicious patterns. It relies on `subprocess.run` to interact with the 'claude' CLI, which is a controlled external dependency. The primary security consideration is that users might store sensitive authentication tokens (bearer, API keys) in plaintext within the `mcp-servers.yaml` file, which is a common but user-managed risk. The script faithfully processes URLs and tokens provided in the configuration, so the security of the configured MCP servers depends on the trustworthiness of those URLs and the handling of the `mcp-servers.yaml` file itself by the user.

Updated: 2026-01-17GitHub→

Medium Cost

mimi-panda-mcp-server

by merdekiti

Sec9

Provides an MCP-compliant interface for AI assistants to access Mimi Panda's image processing, AI image generation, and creative content services.

Setup Requirements

⚠️Requires Node.js 18+.
⚠️Requires external access to the Mimi Panda API (either self-hosted or cloud instance).
⚠️Users must manually obtain a Mimi Panda API token from the Mimi Panda application (web or desktop) for authenticated requests.

Verified SafeView Analysis

The project demonstrates a strong focus on security, as evidenced by its PRODUCTION_CHECKLIST, which includes explicit mentions of input validation (Zod schemas), error handling, timeout management, and sensitive header masking. Communication primarily occurs via STDIO, a secure model for local client-server interaction that reduces external network attack surfaces compared to a publicly exposed web server. No 'eval' or malicious obfuscation was found in the provided code snippets, and sensitive information like API tokens is managed through environment variables or per-request parameters rather than hardcoding.

Updated: 2025-11-24GitHub→

Medium Cost

mcp-agent-base

by gisetia

Sec9

A minimal FastAPI service exposing an OpenAI-compatible /chat/completions endpoint backed by Anthropic Claude with MCP tool calls and SSE streaming.

Setup Requirements

⚠️Requires an Anthropic API Key (paid service) unless running in simulation mode (`SIMULATE=true`).
⚠️Requires an MCP server to be reachable at `MCP_SERVER_URL` for tool discovery and invocation, unless simulation or empty MCP mock is enabled.
⚠️Requires Python 3.10+.

Verified SafeView Analysis

Secrets (API keys, URLs) are handled via environment variables, avoiding hardcoding. The service implements retry mechanisms for network calls to both the LLM provider and the MCP server, improving resilience. No 'eval' or other direct arbitrary code execution patterns were found. Security is partly dependent on the trustworthiness of the external MCP server and the tools it exposes.

Updated: 2026-01-19GitHub→

Low Cost

mcp-auth-go-sdk

by djoreilly

Sec8

Tests OAuth support for the Model Context Protocol (MCP) Go SDK using Keycloak and MCP Inspector, providing an example MCP server with authenticated tools.

Setup Requirements

⚠️Requires a running Keycloak instance, which needs to be configured with a specific realm, client, scopes, and users via a provided Python script and manual UI steps (Web Origins, Valid Redirect URIs).
⚠️Requires `npx @modelcontextprotocol/inspector@0.16.7` as specific older version is needed due to a known bug in newer versions.
⚠️The Keycloak URL and server audience are hardcoded in the Go source, requiring recompilation for changes.

Verified SafeView Analysis

Implements standard OAuth 2.0 Bearer token authentication with JWT validation using Keycloak as the OIDC provider. It fetches and automatically refreshes JWKS for cryptographic key validation, which is good practice. The server binds to localhost, limiting direct external exposure. No obvious malicious patterns, 'eval' usage, or hardcoded sensitive secrets are present for its intended purpose as an example server. Production deployments would typically require additional hardening such as rate limiting (which the code explicitly notes as a potential need).

Updated: 2026-01-08GitHub→

Low Cost

Clearcom-MCP-Server

by RobertCrupa

Sec2

Provides an AI agent with an API to manage ClearCom MCP server connections, roles, and channel assignments.

Setup Requirements

⚠️Requires Python 3.12 or higher.
⚠️Requires the 'uv' package manager for installation and execution.
⚠️Hardcoded ClearCom device IP (http://10.50.16.99) and credentials ('admin'/'admin') in `src/main.py`, which must be changed in the source code to work with a different device or secure credentials.

Review RequiredView Analysis

CRITICAL: The `src/main.py` file hardcodes the `DeviceClient` base_url, username, and password (`admin:admin`). This is a severe security vulnerability as it exposes sensitive access credentials directly in the source code and prevents dynamic configuration for different environments. These values should be loaded from environment variables or a secure configuration mechanism.

Updated: 2025-11-21GitHub→

Medium Cost

monkdb-mcp

by manohar9694

Sec9

Enables LLMs and AI agents to securely interact with MonkDB for data querying, description, and health checks using a standardized Model Context Protocol (MCP).

Setup Requirements

⚠️Requires a running MonkDB instance accessible from the server's host.
⚠️Requires granting only minimum necessary privileges to the database user configured for the MCP server.
⚠️Python 3.13+ for the Python version, or Node.js 20+ and NPM 10.9+ for the TypeScript version.

Verified SafeView Analysis

The server explicitly blocks non-SELECT queries in the `run_select_query` tool, a critical measure against SQL injection for write/delete/update operations. Configuration relies on environment variables (MONKDB_HOST, MONKDB_USER, MONKDB_PASSWORD, MONKDB_API_PORT), preventing hardcoded secrets. Logging is implemented for server operations. A crucial caveat, highlighted in the README, is the user's responsibility to configure the database user with the 'minimum necessary privileges' for the MCP server, avoiding default or administrative users. This external configuration is paramount for the overall security posture.

Updated: 2025-12-15GitHub→

PreviousPage 525 of 760Next