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)

0
0
Medium Cost
matuteduro icon

mcp-server

by matuteduro

Sec9

To serve a React design system's components and instructions to an LLM via Model Context Protocol (MCP) for guided UI generation.

Setup Requirements

  • ⚠️Requires a 'design-system.json' file to be present in the root directory relative to 'src/mcp-server.js'.
  • ⚠️Requires Node.js runtime.
Verified SafeView Analysis
The server communicates over standard I/O (stdio), significantly reducing its external network attack surface. It reads a local JSON file ('design-system.json') and exposes its content via MCP resources and tools. No 'eval' or explicit dangerous operations detected. Security relies on the integrity of the 'design-system.json' file.
Updated: 2025-11-26GitHub→
0
0
Low Cost
Cloud-Ops-Dev icon

n8n.mcp

by Cloud-Ops-Dev

Sec3

This server is a multi-cloud infrastructure orchestration platform that combines n8n automation with Terraform to build, deploy, and manage distributed applications across AWS, IBM Cloud, and on-premises environments, featuring an AI-assisted workflow creation via MCP.

Setup Requirements

  • ⚠️Requires Docker and Docker Compose to be installed and running.
  • ⚠️Requires AWS and IBM Cloud accounts with appropriate API keys and SSH key pairs configured.
  • ⚠️Requires SSH access to a local AMD workstation and a specific Arch workstation (192.168.1.43) for the full message queue demo.
  • ⚠️Initial setup requires populating sensitive credentials (API keys, SSH paths, passwords) in `docker/.env`.
Review RequiredView Analysis
The `N8N_ENCRYPTION_KEY` which encrypts all credentials within n8n is hardcoded in `docker-compose.yml` and explicitly used in utility scripts. This is a severe vulnerability for any non-lab, production environment, as compromise of this key means all cloud credentials stored in n8n are exposed. Additionally, many deployment scripts and workflows use `ssh -o StrictHostKeyChecking=no`, which disables SSH host key verification, making connections vulnerable to Man-in-the-Middle attacks. Cloud API keys and SSH private keys are stored in plaintext in `.env` files and/or written to `terraform.tfvars` within mounted Docker volumes, making them accessible to the n8n container. While common for automation agents, it necessitates robust security of the host system and the n8n container.
Updated: 2025-12-31GitHub→
0
0
Low Cost
Ethan2298 icon
Sec9

Enables semantic search over local notes and documents, integrated with MCP-compatible clients like Claude Code.

Setup Requirements

  • ⚠️Requires Python 3.10+.
  • ⚠️First startup involves downloading an ~80MB embedding model, taking ~10 seconds.
  • ⚠️Requires ~500MB of disk space for the model and dependencies.
Verified SafeView Analysis
The server appears generally safe. It uses local storage (ChromaDB) and a local embedding model (sentence-transformers), minimizing external network exposure beyond initial model download. File operations are handled using `pathlib` for robustness. Input queries are processed as data for embedding and search, not executed as code. There are no obvious hardcoded secrets, `eval` usage, or direct shell command execution of untrusted input. Potential risks are limited to vulnerabilities within its well-known third-party dependencies (e.g., `sentence-transformers`, `chromadb`, `beautifulsoup4`, `PyMuPDF`).
Updated: 2025-12-14GitHub→
0
0
Medium Cost
ForestAdmin icon

mcp-server-metabase

by ForestAdmin

Sec2

Connects AI agents to Metabase BI platform, providing comprehensive access to databases, saved questions, dashboards, collections, and query execution.

Setup Requirements

  • ⚠️Requires access to a Metabase instance (self-hosted or cloud).
  • ⚠️Requires a Metabase API key or username/password to connect to Metabase.
  • ⚠️Requires a manually generated MCP_AUTH_TOKEN for server authentication.
Review RequiredView Analysis
CRITICAL SQL INJECTION VULNERABILITIES (CVE-2025-XXXXX) have been identified and are present in the provided source code. The `execute_sql_query` tool allows `WITH` (Common Table Expression) queries which can contain arbitrary SQL statements (UPDATE, INSERT, DELETE, DROP), bypassing read-only validation. The `create_question` tool lacks any input validation for its `query` argument, allowing malicious SQL to be saved as persistent questions. These vulnerabilities enable data modification, destruction, and privilege escalation, as detailed in `CRITICAL_SECURITY_ADVISORY.md`. Despite previous security patches, these critical issues are unaddressed in the provided code.
Updated: 2025-12-03GitHub→
0
0
Medium Cost
ach968 icon
Sec9

Aggregates and provides search capabilities for a user's saved posts/bookmarks from Reddit and X (Twitter) via MCP tools.

Setup Requirements

  • ⚠️Requires exporting browser cookies for Reddit and X (Twitter) and providing them via environment variables (X_COOKIES_FILE/X_COOKIES, REDDIT_COOKIES_FILE/REDDIT_COOKIES) or default paths.
  • ⚠️Requires setting the REDDIT_USERNAME environment variable for Reddit scraping.
  • ⚠️Requires Playwright browser binaries to be installed (`uv run playwright install chromium`).
  • ⚠️Requires Python 3.13 or higher.
  • ⚠️If running via Docker, Docker daemon must be running.
Verified SafeView Analysis
The server uses Playwright for web scraping, which involves launching a browser. Sensitive user data (authentication cookies and Reddit username) are expected to be provided via environment variables or files, not hardcoded. There are no 'eval' statements or obfuscation found. The risk largely depends on the security of the provided cookies and the environment it runs in.
Updated: 2026-01-17GitHub→
0
0
Medium Cost

tushare-mcp-server

by guojianchuan821113

Sec8

Integrate Tushare Pro financial data APIs into AI clients (e.g., Claude Desktop) via a local MCP server.

Setup Requirements

  • ⚠️Requires a valid Tushare Pro Token (and likely an account/subscription with Tushare Pro).
  • ⚠️Python 3.10+ is a strict requirement.
  • ⚠️The `uv` dependency manager must be installed globally or accessible in the environment.
Verified SafeView Analysis
The server primarily acts as a local `stdio` wrapper for the Tushare Pro API. It requires a Tushare Pro Token, which should be handled securely (e.g., via environment variables or .env file) to prevent unauthorized access to Tushare Pro data. There are no apparent 'eval' or obfuscation risks. Outbound network calls are made to Tushare Pro's official API.
Updated: 2025-11-17GitHub→
0
0
Low Cost

mcp-server-weather

by truck-intel

Sec9

A simple Model Context Protocol (MCP) server that provides static weather information.

Verified SafeView Analysis
The server is a demo implementation with static weather data, meaning it does not interact with external APIs or handle sensitive dynamic data. No indicators of 'eval' or obfuscation are present in the provided information. It likely opens a network port, typical for a server, but with minimal inherent risk given its static nature.
Updated: 2025-11-20GitHub→
0
0
High Cost
nomad3 icon

techopsmind

by nomad3

Sec8

Company-agnostic SRE operations platform with AI-powered monitoring, plugin architecture for multi-company support, and comprehensive infrastructure observability via Model Context Protocol (MCP). It acts as an AI-powered operations assistant for SRE teams.

Setup Requirements

  • ⚠️Requires GEMINI_API_KEY (paid Google service)
  • ⚠️Requires VPN connection for Prometheus and SSH access to internal infrastructure
  • ⚠️Requires Docker and Docker Compose (for full web stack deployment)
  • ⚠️Requires Python 3.10+ (managed by Poetry) and Go 1.21+ (for optional TUI dashboard), Node.js/npm (for React frontend)
  • ⚠️High resource requirements: ~2GB disk space for vector store, ~750MB-1.5GB memory for the MCP server process
  • ⚠️SSH multi-hop access requires specific configuration (`SSH_GATEWAY_PASSWORD` and potentially legacy SSH algorithms/host key management for older components)
Verified SafeView Analysis
The server incorporates strong security practices for an internal SRE tool. It leverages environment variables for all sensitive credentials (API keys, tokens, passwords). Input validation, command sanitization, and path traversal prevention are implemented via `core/security/validators.py`. The full web stack deploys with Docker containers for isolation and Nginx/Certbot for HTTPS/SSL. Access to internal Prometheus and SSH infrastructure requires an active VPN. While older documentation mentioned a 'StrictHostKeyChecking no' workaround for SSH multi-hop authentication, the refactored plugin architecture aims for more secure SSH key-based authentication. Operations involving root access on target servers are a high privilege but justified for its SRE purpose. The presence of robust input validators helps mitigate command injection risks.
Updated: 2025-12-03GitHub→
0
0
Low Cost
tyranosurasmax icon

MCP-CODEGEN

by tyranosurasmax

Sec3

Generates type-safe TypeScript wrappers for MCP, REST, and GraphQL APIs to enable AI agents to interact with external services with 98% token reduction.

Setup Requirements

  • ⚠️Requires Node.js version 18 or higher.
  • ⚠️MCP server adapter has known schema validation issues that prevent tool discovery for MCP sources in current version.
  • ⚠️Execution of commands for MCP servers (via `npx` or custom binaries) runs with full system privileges, posing a security risk if configuration files are compromised or not rigorously controlled.
Review RequiredView Analysis
The MCPAdapter and RawMCPClient components use Node.js `child_process.spawn` to execute commands and arguments defined in configuration files (e.g., `command: 'npx'`, `args: [...]`). If an attacker can inject or modify a configuration file, they could execute arbitrary code on the host system. The project explicitly states, 'No sandboxing: Generated code runs with full privileges' and lists sandboxing as a planned feature, indicating a known critical risk. While authentication for external APIs uses environment variables, the ability to run arbitrary child processes from configuration is a significant vulnerability for an AI agent tool.
Updated: 2026-01-06GitHub→
0
0
Low Cost
rousseya icon
Sec7

Provides an MCP server to interact with Strava, offering activity retrieval, performance statistics, and activity management tools like renaming and e-bike detection.

Setup Requirements

  • ⚠️Requires creating a Strava App and obtaining OAuth tokens (multi-step process involving browser authorization).
  • ⚠️Requires `uv` for dependency management.
  • ⚠️Python 3.10+ is required.
Verified SafeView Analysis
The `scripts/get_tokens.py` file hardcodes `CLIENT_ID` and `CLIENT_SECRET`. While these appear to be public example credentials often used in Strava API examples and not a leak of private keys, hardcoding credentials in source code is a significant security anti-pattern and should be avoided. The server's main application code (`app.py`, `strava_client.py`) correctly utilizes environment variables for all secrets. `TrustedHostMiddleware` in `app.py` is configured with `allowed_hosts=["*"]`, which is overly permissive for general production use but might be acceptable within a managed container environment like Hugging Face Spaces; for other deployments, it should be restricted. The optional `API_TOKEN` for bearer authentication to the MCP endpoint is a good security feature.
Updated: 2025-12-18GitHub→
0
0
Medium Cost
lokesh-reddy-afb icon

mcp-server

by lokesh-reddy-afb

Sec3

Provides tools to search files and directories using regular expressions, returning matched lines and line numbers.

Setup Requirements

  • ⚠️Requires Node.js >= 18.0.0
  • ⚠️Requires file system access permissions to specified paths
Review RequiredView Analysis
Critical vulnerabilities identified: User-supplied regex patterns are used directly to construct `new RegExp()`, leading to potential Regular Expression Denial of Service (ReDoS) attacks. User-supplied file and directory paths are used with `fs.readFileSync` and `fs.readdirSync`, which, despite using `path.resolve`, could potentially lead to arbitrary file disclosure or traversal if not properly sandboxed or if the process has elevated permissions to sensitive areas.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
jordlee icon
Sec9

A RAG-based MCP server providing semantic search and information retrieval from Sony camera help guides, enabling LLMs to answer camera-related queries and perform comparisons.

Setup Requirements

  • ⚠️Requires Pinecone API Key (Paid Service) for functionality.
  • ⚠️Requires a pre-populated Pinecone index (e.g., 'camera-rag-agent') with processed camera help guide data, involving several preprocessing scripts (download, parse, chunk, embed).
  • ⚠️Requires Redis for rate-limiting functionality (optional for basic operation, but crucial for production stability).
Verified SafeView Analysis
The server correctly uses environment variables for sensitive API keys like PINECONE_API_KEY. Input validation is handled implicitly by passing parameters to the Pinecone client's query filters, mitigating direct injection risks in the search logic. The rate limiter adds a layer of protection against abuse. No direct `eval` or `os.system` calls were found with user-controlled input. However, the repository contains a `scripts/clear_pinecone_index.py` utility that can delete all vectors from the Pinecone index; while it includes a user confirmation step, it highlights a powerful and potentially destructive capability in the codebase. The `mcp/chatgpt.md` file also explicitly outlines the broader LLM-level prompt injection and data exfiltration risks when *using* such a server with an AI model, which are inherent to agentic systems and not solely code-level vulnerabilities within the server itself.
Updated: 2025-12-05GitHub→
PreviousPage 418 of 760Next