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
muhammadhayat123 icon

MCP_client_server

by muhammadhayat123

Sec9

This project demonstrates client-server delegation of LLM tasks using the MCP framework, where the server requests an LLM generation from the client.

Setup Requirements

  • ⚠️Python 3.12+ required
  • ⚠️MCP library must be installed (pip install mcp)
  • ⚠️Uvicorn server must be installed to run the ASGI application (pip install uvicorn)
Verified SafeView Analysis
The provided code snippet does not contain 'eval', 'exec', or hardcoded secrets. Input 'topic' is embedded as text content and not directly executed. The server runs on localhost, and the security relies heavily on the underlying 'mcp' library and proper configuration in a production environment. For a demo, it appears safe.
Updated: 2025-11-19GitHub→
0
0
Low Cost
Sec10

This server provides tools to search and retrieve Swedish library statistics from the National Library of Sweden's open data API.

Setup Requirements

  • ⚠️Requires Node.js environment.
Verified SafeView Analysis
The code does not use 'eval' or similar dangerous functions. It relies on the well-established '@modelcontextprotocol/sdk' and standard Node.js/fetch APIs. No hardcoded secrets are present. Network requests are made to a public API endpoint. The code appears safe to run.
Updated: 2025-11-21GitHub→
0
0
High Cost
maqsam22 icon

claude-code

by maqsam22

Sec9

A comprehensive developer tool that integrates AI agents for structured feature development, git workflow automation, code quality review, and Agent SDK application development and verification.

Setup Requirements

  • ⚠️Requires an Anthropic API Key (Paid) for AI model access.
  • ⚠️Requires GitHub CLI (`gh`) installed and authenticated, along with a `GITHUB_TOKEN` environment variable for full Git workflow automation and issue management scripts.
  • ⚠️The project's scripts are written in TypeScript and use Bun (or Node.js with ts-node) as an execution environment.
Verified SafeView Analysis
The project includes explicit security features like a `security_reminder_hook.py` that warns about common vulnerabilities (e.g., `eval`, `child_process.exec`, XSS patterns) when code is modified. Several AI agents (`code-reviewer`, `silent-failure-hunter`, `agent-sdk-verifier-py/ts`) are also tasked with identifying security issues and bad practices like hardcoded API keys. While the nature of a coding agent means it could be instructed to execute arbitrary commands, the project demonstrates a strong focus on internal security and user guidance to prevent vulnerabilities. The GitHub interaction scripts use environment variables for sensitive tokens.
Updated: 2026-01-19GitHub→
0
0
High Cost
marc-shade icon

synthetic-data-mcp

by marc-shade

Sec9

Generates privacy-compliant synthetic data for training and testing, specifically for regulated industries like healthcare and finance. It supports multiple LLM providers, advanced privacy controls, and various compliance frameworks (HIPAA, GDPR, PCI DSS, SOX).

Setup Requirements

  • ⚠️Requires Python 3.10+.
  • ⚠️For local LLM inference (Ollama), Ollama must be installed and running locally, with the desired models pulled.
  • ⚠️For cloud LLM inference, API keys for providers like OpenAI, Anthropic, Google, or OpenRouter are required.
  • ⚠️The UI requires `npm install` and `npm run dev` in the `ui/` directory if the frontend is to be run locally alongside the backend.
  • ⚠️Requires a backend database for internal operations (e.g., PostgreSQL in production, SQLite for local dev).
Verified SafeView Analysis
The system demonstrates a strong focus on security and privacy through robust architectural components like `PrivacyEngine`, `ComplianceValidator`, `AuthService` (using `CryptContext` for hashing, JWT), `DataEncryptionKeyManager` (supporting multiple KMS), and `AuditTrail`. It explicitly guarantees no retention/output of original PII. Environment variables are correctly used for API keys. The presence of demo credentials in `LoginPage.tsx` is noted for development/testing, but is clearly not for production use. No 'eval' or direct malicious patterns were detected. Deployment manifests for Kubernetes and Helm indicate a production-ready security posture.
Updated: 2025-12-29GitHub→
0
0
Medium Cost
08car1118 icon
Sec8

Provides RAG search capabilities using the Gemini File Search API for AI assistants to manage and query documents.

Setup Requirements

  • ⚠️Requires a Google Gemini API Key, which incurs costs for document indexing ($0.15 / 1M tokens) and for the underlying LLM inference during RAG search (Gemini model costs, not explicitly detailed in README's cost section).
  • ⚠️Requires Python 3.11 or newer.
  • ⚠️File upload (`gemini_upload_file`) expects local file paths accessible by the MCP server, not remote URLs.
Verified SafeView Analysis
The `gemini_upload_file` tool takes a `file_path` parameter, which the server reads from the local filesystem. While the `_validate_file` method performs checks for existence, size, and MIME type, exposing this tool to untrusted users could potentially allow attempts at path traversal or access to unintended local files. However, within the context of an MCP server where the caller is a trusted AI assistant, this is a standard and expected pattern for file processing. No direct `eval` calls or unsanitized shell commands were found. API keys are managed via environment variables.
Updated: 2025-12-02GitHub→
0
0
Low Cost
orpheliedomma-a11y icon

MCP-Server-Project

by orpheliedomma-a11y

Sec9

This server provides a Micro-Capability Platform (MCP) for AI agents to plan travel itineraries, fetch weather forecasts, and recommend hotels for a given destination.

Setup Requirements

  • ⚠️Requires `OPENWEATHER_API_KEY` to be set as an environment variable.
  • ⚠️Designed to run with Claude Desktop or a client that connects via standard I/O (stdio) transport.
Verified SafeView Analysis
Uses environment variables for API keys (OPENWEATHER_API_KEY), reducing the risk of hardcoded secrets. No 'eval' or obvious obfuscation found. External API calls are made to OpenWeatherMap, which is a standard practice and not inherently a risk from the server's code perspective.
Updated: 2025-11-25GitHub→
0
0
Medium Cost
vyogotech icon

frappe_ai

by vyogotech

Sec8

Provides an AI assistant for Frappe/ERPNext users by integrating with an external Model Context Protocol (MCP) server for data-aware queries.

Setup Requirements

  • ⚠️Requires a separate, running MCP Server instance, configured to trust the Frappe instance's OAuth2 provider.
  • ⚠️Manual or script-assisted OAuth2 Client setup is required within Frappe, followed by copying Client ID and Secret into MCP Server Settings.
  • ⚠️The MCP Server's `config.yaml` must be updated to specify Frappe's OAuth2 endpoints and trusted clients.
Verified SafeView Analysis
Uses standard OAuth2 client credentials grant for app-to-app authentication, with the client secret stored as a Frappe 'Password' field (implying encryption). User context is transmitted via Frappe session ('sid' cookie) to the MCP server, requiring the MCP server to be configured to trust and validate Frappe's sessions. Configurable network timeouts are implemented. Token caching is in-memory by default, with a recommendation for Redis in production. No 'eval' or other high-risk dynamic code execution patterns from user input were found. The overall security relies heavily on the correct and secure configuration of both the Frappe instance and the external MCP server.
Updated: 2025-11-27GitHub→
0
0
High Cost
sachdev27 icon
Sec7

Enables AI assistants (like Claude) to interact with Dell PowerScale (Isilon) storage clusters by dynamically generating tools from the OneFS REST API specification.

Setup Requirements

  • ⚠️Requires Python 3.10 or higher.
  • ⚠️Requires access to a Dell PowerScale cluster with its REST API enabled and appropriate API credentials.
  • ⚠️Requires a local OpenAPI specification file (e.g., `powerscale_9.7_comprehensive_openapi.json`) from which to generate tools. This file is expected to be present in the installation or specified via `LOCAL_OPENAPI_SPEC_PATH`.
Verified SafeView Analysis
The server uses HTTP Basic Authentication, which is sensitive. While credentials are to be provided via environment variables or per-request arguments (not hardcoded), the default `ISILON_TLS_VERIFY` setting for PowerScale API connections is `false`, which disables SSL certificate validation and poses a significant security risk for data in transit. The HTTP/SSE server (for n8n/web clients) uses `Access-Control-Allow-Origin: *` (open CORS), which is acceptable for internal/controlled environments but dangerous if exposed to the public internet. No instances of `eval` or obvious malicious patterns were found. The codebase appears professionally structured with explicit error handling.
Updated: 2025-12-10GitHub→
0
0
Low Cost
VenkatRamReddyK icon

git-repos-mcp

by VenkatRamReddyK

Sec9

This server exposes a Model Context Protocol (MCP) resource to list the public GitHub repositories of a specific user.

Setup Requirements

  • ⚠️Requires Node.js installed.
  • ⚠️For Node.js versions older than 18, `node-fetch` must be manually installed and polyfilled globally.
Verified SafeView Analysis
The server makes an API call to a standard GitHub endpoint. No 'eval' or obfuscation is present. There are no hardcoded secrets, and the username for the GitHub API call is public information. The network risk is minimal, relying on the security of the GitHub API itself.
Updated: 2025-11-24GitHub→
0
0
Medium Cost
dyl104 icon
Sec9

Provides a FastMCP-based server with basic arithmetic tools, dynamic resources, and sophisticated AI prompt generation templates.

Setup Requirements

  • ⚠️Python 3.13+ required
  • ⚠️Requires `mcp[cli]` installation
Verified SafeView Analysis
The provided source code for `main.py` does not contain obvious security vulnerabilities such as `eval`, hardcoded credentials, or risky file operations. The server exposes tools and prompts, and their safety depends on the inputs provided to them and the underlying `mcp` framework. Input validation for `name`, `task`, `raw_prompt`, etc., is implicit to the tool/prompt function signatures and not explicitly shown, but typically handled by the framework.
Updated: 2025-11-24GitHub→
0
0
High Cost
DawidSu icon

ha-mcp-server

by DawidSu

Sec5

Provides a Model Context Protocol (MCP) server for Claude AI to securely and efficiently interact with Home Assistant configuration files, offering both Docker-based server deployment and Linux desktop integration tools.

Setup Requirements

  • ⚠️Requires Docker and Docker Compose for server deployment.
  • ⚠️The Home Assistant Claude MCP Server Addon must be running in Home Assistant for core functionality and desktop tool integration.
  • ⚠️Linux desktop users leveraging the desktop tools need to perform SSH key setup (ssh-keygen, ssh-copy-id) and install Python 3.6+ with tkinter and netcat.
Verified SafeView Analysis
The project demonstrates significant effort in security, implementing input validation, file access control (path traversal, extension/size limits, suspicious patterns), and rate limiting via `security-utils.sh`. It uses non-root Docker users and includes extensive security documentation. However, the `cache-manager.sh` script uses `eval "$command"`, allowing arbitrary command execution. While this is currently used for internal commands, its presence in a script callable by the Node.js API server (via `child_process.exec`) poses a critical remote code execution (RCE) vulnerability if an injection vector were to be discovered or introduced. Similarly, the `claude-mcp-widget.py` uses `subprocess.Popen` with f-strings for SSH commands, which, if not properly sanitized or if user input is maliciously crafted, could lead to command injection, though SSH clients typically offer some protection. Users must ensure strict control over all inputs and environments.
Updated: 2026-01-19GitHub→
0
0
High Cost
saurabhmain icon

a-mem-mcp-server

by saurabhmain

Sec5

An agentic memory system for LLM agents, enhancing research by storing, linking, and evolving knowledge based on the Zettelkasten principle, with IDE integration.

Setup Requirements

  • ⚠️Requires a Python 3.x environment with various dependencies (e.g., `textual`, `requests`, `dash`, `plotly`).
  • ⚠️If using Ollama, requires a local Ollama server running and specific models (`qwen3:4b`, `nomic-embed-text`) to be pulled.
  • ⚠️If using OpenRouter, requires a paid API key (`OPENROUTER_API_KEY`).
  • ⚠️For advanced graph backends like FalkorDB, specific installations are needed (`falkordblite` for Linux/macOS, or `falkordb` and `redis` with the FalkorDB module for Windows).
  • ⚠️For web research and PDF extraction, may require local Docker containers for Jina Reader or Unstructured API, or their respective Python libraries and dependencies like `pdfminer.six`.
Review RequiredView Analysis
The server exposes a `/get_graph` endpoint via an `aiohttp` web server, by default on `127.0.0.1:42424`, which is read-only but exposes internal graph data. The `add_file` tool processes file paths provided by the user/agent; if a malicious actor controls this input, it could potentially be used to read arbitrary files accessible to the server process, leading to information disclosure. Additionally, public Google API keys are hardcoded as defaults, which, while likely for a public Custom Search Engine, is still poor practice and could lead to API quota exhaustion.
Updated: 2026-01-19GitHub→
PreviousPage 608 of 713Next