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(8554)

0
0
Low Cost
delonsp icon

rlm-mcp-server

by delonsp

Sec8

Manages large datasets outside of an LLM's context, providing a persistent Python REPL and tools for data analysis, PDF processing, and S3 integration to enable Recursive Language Models.

Setup Requirements

  • ⚠️Requires Docker for server deployment.
  • ⚠️Requires OPENAI_API_KEY for recursive LLM sub-calls (paid service).
  • ⚠️Requires MISTRAL_API_KEY for PDF OCR (paid service).
  • ⚠️Requires SSH tunnel or public HTTPS endpoint configured for client connection (socat for SSH tunnel option).
  • ⚠️Python 3.11+ is required.
Verified SafeView Analysis
The server employs a sandboxed Python REPL using `exec()`, but mitigates risks with `ast.parse` for static code analysis, whitelisted imports (`ALLOWED_IMPORTS`), and blocked built-in functions (`BLOCKED_BUILTINS`). File access is restricted to a read-only `/data/` volume with path traversal checks. Environment variables are used for API keys. However, the `README.md` misleadingly states 'Container em rede isolada (sem acesso à internet)' under security, while `llm_client`, `s3_client`, and `pdf_parser` (Mistral OCR) components explicitly require external network access. While the internal sandbox is robust for the intended use, network access to external APIs should be clearly stated as a dependency.
Updated: 2026-01-19GitHub→
0
0
Low Cost
sanjay3290 icon

hasura-mcp-server

by sanjay3290

Sec8

Enables AI assistants to interact with and manage Hasura GraphQL Engine instances through natural language.

Setup Requirements

  • ⚠️Docker required
  • ⚠️Requires a running Hasura GraphQL Engine instance
  • ⚠️Requires Hasura admin secret (recommended for full functionality)
  • ⚠️Requires an MCP-compatible AI client (e.g., Claude Desktop) for interaction
Verified SafeView Analysis
Secrets (Hasura admin secret) are handled via command-line flags or environment variables, not hardcoded within the provided code. The server can be run in a read-only mode for production safety, disabling destructive operations. It leverages Docker for containerization, providing isolation. The source code is built upon the Hasura GraphQL Engine's CLI internal libraries and the MCP-Go SDK, which are established open-source projects. The use of `--network host` for local instances is a configuration detail for local networking rather than a general vulnerability, but should be understood for its implications.
Updated: 2025-11-24GitHub→
0
0
Low Cost
ML-and-AI icon
Sec3

Implements a single-server solution for the Minecraft Communication Protocol to interact with Minecraft clients or servers.

Review RequiredView Analysis
Insufficient source code provided for a meaningful security audit. Cannot verify for 'eval', obfuscation, network risks, hardcoded secrets, or malicious patterns. Running unverified code carries inherent risks.
Updated: 2025-12-11GitHub→
0
0
High Cost
kms12425-ctrl icon

mcp-server-project

by kms12425-ctrl

Sec4

An AI agent-based Micro-Agent Protocol (MCP) server designed for generating various types of reports, including technical solutions, experimental reports, review reports, analysis reports, and patents, utilizing a 'DeepResearch' mode.

Setup Requirements

  • ⚠️Requires Python >= 3.10.
  • ⚠️Requires the 'uv' package manager to be installed for dependency synchronization (`uv sync`).
  • ⚠️Requires `sops` and `age` encryption tools to be installed for managing encrypted secrets (e.g., `env.yaml`). Setup scripts are provided for Linux/macOS.
  • ⚠️Cloning the repository with submodules from `ssh://git@222.20.98.63:3001/YA/YA_MCPServer_DeepReport.git` may require specific SSH configurations or internal network access.
Review RequiredView Analysis
The server uses `subprocess.run` for `git` commands and `sops` decryption, which requires careful input sanitization to prevent command injection vulnerabilities. While secrets are managed securely using `sops` and `age` encryption, the SSE transport implementation includes a `CORSMiddleware` with `allow_origins=["*"]`. This permissive CORS policy is a significant security risk for any non-local deployment, potentially enabling Cross-Site Request Forgery (CSRF) and other unauthorized access from arbitrary domains.
Updated: 2026-01-18GitHub→
0
0
Low Cost
waalexan icon

mcp_server

by waalexan

Sec8

This server provides an API for AI agents and LLMs to interact with web application functionality and retrieve content, adhering to the Model Context Protocol (MCP) using Next.js.

Setup Requirements

  • ⚠️Requires `NEXT_PUBLIC_BASE_URL` environment variable to be set for proper functioning outside of Vercel or local development defaults.
  • ⚠️Requires Node.js version compatible with Next.js 16.0.6 (typically >=20.9.0).
  • ⚠️Designed to integrate with an external Model Context Protocol (MCP) client/LLM, not a standalone web application.
Verified SafeView Analysis
The codebase generally follows good security practices for a Next.js application. Input validation for registered tools is performed using Zod, mitigating common injection risks. User-provided data (`newValue.name` in `page.tsx`) is rendered within JSX, which automatically escapes string content, preventing Cross-Site Scripting (XSS) in that context. The `getHtmlFromPage` utility fetches content only from the application's own `baseURL`, reducing Server-Side Request Forgery (SSRF) risk. No direct use of `eval` or obvious obfuscation was found. The main security considerations would be the broader ecosystem's handling of the `text/html+skybridge` widget content and the trustworthiness of external LLM interactions.
Updated: 2025-12-19GitHub→
0
0
Medium Cost
rach icon
Sec9

This server provides a structured way for LLM applications to retrieve pre-defined markdown instructions and context, such as project workflows or code review checklists, via the Model Context Protocol (MCP).

Setup Requirements

  • ⚠️Requires Node.js version 18 or higher.
  • ⚠️A `instruction.yaml` configuration file is required, defining tools and their associated markdown file paths.
  • ⚠️All markdown files referenced in `instruction.yaml` must exist at the specified relative paths (relative to the config file).
Verified SafeView Analysis
The server's core functionality involves reading local markdown files based on paths defined in the `instruction.yaml` configuration. Trusting the `instruction.yaml` file is crucial, as it dictates which local files can be served. There are no direct `eval` calls, obfuscation, or hardcoded secrets found. By default, it operates over standard I/O (stdio), limiting direct network exposure unless explicitly configured for HTTP (which is not the default or primary mode shown).
Updated: 2025-12-13GitHub→
0
0
Low Cost
raineislam icon

mcp-framework

by raineislam

Sec6

Remote control, video streaming, and sensor data collection for a Raspberry Pi-based robotic car system.

Setup Requirements

  • ⚠️Requires a Raspberry Pi for client-side scripts (ultrasonic, video streaming, GPIO control).
  • ⚠️Requires a PiCamera module for video streaming functionality.
  • ⚠️Requires an Arduino or similar microcontroller connected via USB serial for car control commands (as seen in `test/car.py`).
  • ⚠️Specific Python libraries (RPi.GPIO, picamera, pygame, curses, serial, numpy, opencv-python) must be installed.
Verified SafeView Analysis
Network communication (sockets) lacks encryption and authentication, making it suitable only for trusted local networks. Hardcoded IP addresses in client scripts are inflexible and could point to unintended destinations if the network configuration changes. No 'eval' or overtly malicious patterns were found.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
pcontezini icon

asaas-mcp-server

by pcontezini

Sec9

This server acts as an MCP (Model Context Protocol) intermediary, dynamically generating tools from the Asaas OpenAPI specification to enable LLMs to interact with the Asaas financial API via STDIO/JSON-RPC.

Setup Requirements

  • ⚠️Requires Node.js 24+.
  • ⚠️Requires 'SANDBOX_API_KEY' or 'ASAAS_API_KEY' environment variable for authentication.
  • ⚠️OpenAPI spec is dynamically fetched from Asaas.com (cached locally), requiring internet access on first run or specifying 'ASAAS_SPEC_PATH' for offline use.
Verified SafeView Analysis
The server primarily relies on environment variables for sensitive data like API keys, which is a good practice. It explicitly checks for the presence of an API key. HTTP requests are made over HTTPS, and basic retry logic for rate limits/network errors is included. Path parameters are URL-encoded, preventing path traversal. Logging can be directed to a file but is carefully handled to prevent crashes. There are no obvious signs of eval() usage or code obfuscation. The server acts as a proxy, so its security profile is largely dependent on the underlying Asaas API's security. Input schemas for tools are derived from the OpenAPI spec, providing a level of input validation.
Updated: 2025-12-02GitHub→
0
0
High Cost
AsobaCloud icon

asobacode.github.io

by AsobaCloud

Sec6

AI-powered energy asset management, forecasting, and operations & maintenance (O&M) automation for solar energy assets.

Setup Requirements

  • ⚠️Requires Python 3.10+ (explicitly stating not to use system Python 3.9).
  • ⚠️Requires an AWS account and AWS CLI configured with specific IAM permissions for services like EC2, Lambda, API Gateway, S3, DynamoDB, CloudWatch, and Bedrock.
  • ⚠️Optimal performance for LLM inference (e.g., PolicyAnalyst) necessitates a GPU instance (e.g., AWS g5.2xlarge or g4dn.xlarge), incurring significant cloud costs.
  • ⚠️Requires a Hugging Face token (HF_TOKEN) for downloading and serving models.
  • ⚠️Docker is essential for running local LLM inference servers (Hugging Face TGI or vLLM).
Verified SafeView Analysis
The project documentation acknowledges existing vulnerabilities and outstanding security epics (e.g., 'Remove all hardcoded credentials', 'Fix 14 dependency vulnerabilities', and the need to complete 'security and production infrastructure epics first'). While comprehensive security measures are described as being implemented or in progress (secrets management, IAM roles, TLS, encryption, RBAC, audit logs, SOC 2 Type II in progress, GDPR), the explicit mention of known issues and the use of API keys in query parameters (less secure) indicate that security is an ongoing concern and not yet fully hardened for all aspects.
Updated: 2026-01-18GitHub→
0
0
Low Cost
santhosh-genai icon

test-data-mcp-server

by santhosh-genai

Sec9

This MCP server provides tools for generating synthetic test data and fetching prospect leads from an Opentaps MySQL database for QA and development purposes.

Setup Requirements

  • ⚠️Requires a MySQL database (e.g., Opentaps CRM schema) for the `getLatestProspects` tool.
  • ⚠️Requires Node.js version 18.0.0 or higher.
  • ⚠️Requires `npm install` to set up project dependencies.
Verified SafeView Analysis
The server demonstrates good security practices: - Database credentials are loaded from environment variables. - SQL queries use prepared statements, preventing SQL injection for parameterized values. - Input validation is explicitly performed on all tool parameters, checking types, ranges, and allowed enum values. - There is no use of `eval` or dynamic code execution from external input. - Error handling is structured, providing clear JSON-RPC error responses and tool-specific error payloads. - The server operates over `stdio` which is an IPC mechanism, reducing direct network exposure. One point deduction as it's not using an ORM which would fully abstract SQL, but the current approach with prepared statements is robust.
Updated: 2025-12-14GitHub→
0
0
High Cost
qontinui icon

qontinui-web-mcp

by qontinui

Sec8

Enables AI assistants to create and manage visual automation configurations through the Qontinui API for UI automation.

Setup Requirements

  • ⚠️Requires Python 3.12 or higher.
  • ⚠️Requires connectivity to an external Qontinui API backend (local or production).
  • ⚠️Requires authentication via QONTINUI_ACCESS_TOKEN or QONTINUI_EMAIL/QONTINUI_PASSWORD for most operations.
Verified SafeView Analysis
The server acts as a client to an external Qontinui API, facilitating data management (CRUD for projects, workflows, states, etc.) and triggering actions on the backend. It uses standard secure practices like environment variables for sensitive data (API keys, credentials) and Pydantic for data validation. No direct 'eval' or obvious code injection vulnerabilities are present within the MCP server's logic for processing tool arguments. The primary security considerations would shift to the Qontinui backend API itself, which this MCP server interacts with, and how it handles and executes user-defined workflows/actions.
Updated: 2025-12-15GitHub→
0
0
Medium Cost
mathisarends icon

spotify-mcp

by mathisarends

Sec9

An async server for controlling Spotify playback and search functionality via an MCP interface, primarily for agent integration.

Setup Requirements

  • ⚠️Requires Python 3.14+
  • ⚠️Requires Spotify API credentials (Client ID, Secret, Redirect URI) obtained from Spotify Developer Dashboard.
  • ⚠️The 'uv' package is used for running, which may require a separate installation.
  • ⚠️The default SPOTIFY_REDIRECT_URI (`http://localhost:8888/callback`) requires port 8888 to be available.
Verified SafeView Analysis
Credentials (Client ID, Secret, Redirect URI) are correctly loaded from environment variables. The server uses an async wrapper around Spotipy, ensuring synchronous operations don't block. The MCP server communication method (Stdio in examples, FastMCP implies HTTP) requires proper setup to avoid unintended network exposure, but no direct vulnerabilities are observed in the provided code.
Updated: 2026-01-18GitHub→
PreviousPage 620 of 713Next