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
Low Cost
TheseHandsAreSpiders icon

remote-mcp-server-authless

by TheseHandsAreSpiders

Sec8

Deploys a remote Model Context Protocol (MCP) server without authentication on Cloudflare Workers, providing basic calculator tools for AI agents.

Setup Requirements

  • ⚠️Requires a Cloudflare Workers account for deployment.
  • ⚠️Requires Node.js and npm/yarn for local development and dependency management.
  • ⚠️The server is intentionally unauthenticated; consider this carefully if extending for non-demo use cases.
Verified SafeView Analysis
The server is explicitly designed to be 'authless' for demonstration and playground purposes, as indicated in its name and README. This means anyone with the URL can access and use the exposed arithmetic tools. While there are no obvious code vulnerabilities like 'eval' or hardcoded secrets in the provided source, deploying an unauthenticated server in a production environment or with sensitive tools would be a significant security risk.
Updated: 2025-12-12GitHub→
0
0
High Cost
GSA-TTS icon
Sec8

This server acts as a Model Context Protocol (MCP) server, allowing LLMs and other clients to query and retrieve information from the NIH RePORTER grants database.

Setup Requirements

  • ⚠️Requires Python 3.13+
  • ⚠️Relies on external NIH RePORTER API for data access
  • ⚠️Uses `uv` runner (e.g., `uv run`)
Verified SafeView Analysis
The code uses Pydantic for input validation and makes requests to a known NIH API. No 'eval' or hardcoded secrets were found in the provided snippets. The project explicitly states it's a 'proof of concept and is not intended for production use,' suggesting it may lack comprehensive security hardening for production environments, but the visible code itself does not present immediate critical vulnerabilities.
Updated: 2026-01-16GitHub→
0
0
Medium Cost
ApreisP4x icon

perspectives

by ApreisP4x

Sec9

A basic implementation of persistent memory using a local knowledge graph, allowing an AI assistant to remember information across chats.

Setup Requirements

  • ⚠️Requires Node.js and npm/npx to run.
  • ⚠️Persistence of the knowledge graph requires proper volume or path configuration for 'memory.jsonl', as data stored in ephemeral environments will be lost on restart (e.g., Docker without volume mounts, 'npx' in a temporary directory).
  • ⚠️Performs a one-time migration from 'memory.json' to 'memory.jsonl' if a legacy file is detected.
Verified SafeView Analysis
The server operates locally via standard I/O and uses a local JSONL file for persistence. File operations are limited to the configured 'memory.jsonl' path. No 'eval', obfuscation, or direct network exposure. The primary security consideration is ensuring 'MEMORY_FILE_PATH' is configured by the user to a non-sensitive location, but this is a user configuration risk, not an inherent server vulnerability.
Updated: 2025-12-02GitHub→
0
0
Low Cost
adamaclp92 icon

mortgage-mcp-server

by adamaclp92

Sec8

A Node.js MCP server demonstrating a mock mortgage loan application process using UI widgets and protocol tools.

Setup Requirements

  • ⚠️Uses mock data only; not connected to real banking systems.
  • ⚠️Requires Node.js (ESM) and npm to run.
Verified SafeView Analysis
The server uses Zod for schema validation, loads environment variables with dotenv, and defines CSP headers for widgets, which are good practices. There's no evident use of 'eval' or direct command execution based on user input. File system access is limited to loading static HTML assets from a known directory. Input schemas for MCP tools sometimes use `z.any()`, which requires downstream components to handle data securely, but within the server itself, data processing appears safe. The system is explicitly stated to use mock data and not connect to real banking systems, inherently limiting the scope of certain financial security risks.
Updated: 2025-12-13GitHub→
0
0
Medium Cost
misho85 icon
Sec8

Provides remote Docker container log and statistics access via SSH, enabling monitoring and debugging of services.

Setup Requirements

  • ⚠️Requires `Bun` runtime to be installed locally.
  • ⚠️Requires `Docker` to be installed on the remote server(s).
  • ⚠️Requires `SSH access` to remote server(s) configured in `~/.ssh/config`.
Verified SafeView Analysis
Input parameters (container names, timestamps, grep patterns, host aliases) are sanitized using regular expressions and character replacement to mitigate command injection risks before executing SSH commands. The server uses the system's SSH client (`ssh`) with `BatchMode=yes` and `ConnectTimeout=10` options. No 'eval' or obvious hardcoded secrets were found in the provided code snippets. Security relies on proper SSH key management and configuration on the host machine.
Updated: 2026-01-17GitHub→
0
0
Medium Cost
TAMU-AESL icon

NRC-ADAMS-AI

by TAMU-AESL

Sec9

Provides LLMs and AI agents with tools to search, download, and extract text from documents in the NRC ADAMS repository.

Setup Requirements

  • ⚠️Requires Python 3.8 or higher.
  • ⚠️Google API Key and Custom Search Engine ID (GOOGLE_API_KEY, GOOGLE_CX) are optional for hybrid search but necessary for full functionality; Google Custom Search may have usage quotas.
  • ⚠️Requires Docker or Docker Compose for containerized deployment, which is recommended for production and portability.
  • ⚠️Relies on the availability and stability of the external NRC ADAMS API for core functionality.
Verified SafeView Analysis
The server implements good security practices including environment variable usage for secrets (`GOOGLE_API_KEY`, `GOOGLE_CX`), robust path validation for file operations (`summarize_pdf` tool ensures paths are within the designated downloads folder to prevent directory traversal attacks), and proper error handling. No obvious malicious patterns or arbitrary code execution from user input were found. XML parsing from a trusted source (ADAMS API) is generally safe, and input parameters are quoted when building queries. A simple rate limiter is also implemented for API calls.
Updated: 2026-01-06GitHub→
0
0
Medium Cost
Ahmed-M-Elevate icon

elevate-mcp

by Ahmed-M-Elevate

Sec8

Provides an AI-powered interface for querying and analyzing AB Testing data stored in MySQL (PlanetScale) and ClickHouse analytics databases.

Setup Requirements

  • ⚠️Requires Go 1.21+ or a pre-built binary.
  • ⚠️Requires database credentials for both MySQL (PlanetScale) and ClickHouse.
  • ⚠️Specific configuration is needed for various AI clients (e.g., Claude Code, Cursor, GitHub Copilot).
Verified SafeView Analysis
The server exposes database query and execution tools. `execute` and `ch_execute` tools allow arbitrary SQL execution (including DDL/DML), which is a high-risk capability if not secured. The `update_test_status` also permits data modification. However, the project strongly recommends using a 'readonly' toolset for production and provides extensive deployment security guidance, including Nginx basic authentication, IP whitelisting, and rate limiting. Crucially, database credentials are loaded from environment variables and must not be committed to version control. The overall security posture is good, provided the deployment guidelines are strictly followed, especially for access control and read-only mode in production.
Updated: 2025-12-03GitHub→
0
0
Medium Cost
dariuszkowalski-com icon

zai-mcp-proxy-server

by dariuszkowalski-com

Sec9

Provides an MCP proxy server for Z.AI Web Search Prime API, enabling integration with Forge as a standard MCP server.

Setup Requirements

  • ⚠️Requires Z.AI Web Search Prime API key (likely a paid service)
  • ⚠️Requires Node.js 16+
Verified SafeView Analysis
The server uses HTTPS for communication, validates all input parameters with Zod, and handles API keys via environment variables or command-line arguments (not hardcoded). Error handling and timeouts are implemented. There are no direct uses of `eval` or other highly dangerous functions with untrusted input. The server specifically handles double-escaped JSON by unescaping before parsing, which is a safe approach. Minor risk: API key passed via command line arguments can be visible in process listings, though environment variables are also supported and preferred for secrets.
Updated: 2025-12-03GitHub→
0
0
High Cost
thiagofinch icon

BILHON-MCP-ECOSYSTEM

by thiagofinch

Sec2

A comprehensive ecosystem designed to enhance AI agent capabilities through standardized integrations for context management, file system operations, memory, web search, desktop and browser automation, and real-time notifications.

Setup Requirements

  • ⚠️Requires a running N8N instance for workflow automation.
  • ⚠️Requires numerous API keys and environment variables for external services (OpenAI, GitHub, Supabase, ClickUp, Exa, Ntfy, Hetzner, etc.).
  • ⚠️The `desktop-commander` component implies a Windows operating system for its configured paths, despite CI/CD suggesting a Linux environment, indicating potential OS-specific setup complexities.
  • ⚠️The `WF_DNA_INGESTAO_VIDEO` workflow requires `yt-dlp` to be installed on the n8n host and can consume significant local storage for downloaded videos.
Review RequiredView Analysis
The system presents significant security risks. The `desktop-commander` server grants extensive system-level control (file system, processes, clipboard, screenshots, mouse/keyboard control) which, if compromised, allows full control over the host. A critical vulnerability lies in the `WF_DNA_INGESTAO_VIDEO` n8n workflow, where the `executeCommand` node uses `{{ $json.videoUrl }}` directly in a shell command, making it highly susceptible to command injection. This allows an attacker to execute arbitrary commands on the n8n host by crafting a malicious `videoUrl`. Additionally, the reliance on AI for task classification and 'sub-agent' auditing (`WF_SUBAGENTE_AUDITORIA`) introduces risks like prompt injection and AI hallucination, which could lead to bypassing safeguards or unintended actions. Sensitive credentials for numerous third-party services (GitHub, Supabase, ClickUp, Exa, OpenAI, Ntfy, Hetzner, n8n) are required, and their secure management is paramount. Deployment via GitHub Actions uses SSH keys, granting administrative access to servers. Without significant hardening, input validation, and secure operational practices, this system is highly vulnerable.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
Sec9

Provides a local, secure Playwright-based server to enable browser automation for Model Context Protocol (MCP) clients.

Setup Requirements

  • ⚠️Requires Node.js runtime installed locally.
  • ⚠️Playwright browser binaries (e.g., Chromium) must be installed via `npx playwright install chromium`.
  • ⚠️MCP client configuration requires an absolute, platform-specific path to `index.js` which users must adjust for their environment.
Verified SafeView Analysis
The server explicitly states it is 'Tamamen Local', 'Dış bağlantı yapmaz', and 'Sadece local sistemde çalışır', using 'stdio transport' for communication, which significantly reduces network-based attack vectors. It does not require external connections. The primary security risk comes from the actions performed by the browser automation (e.g., navigating to untrusted URLs or executing arbitrary JavaScript via `browser_evaluate`), which could exploit browser vulnerabilities. However, the server's design focuses on local isolation.
Updated: 2025-11-17GitHub→
0
0
Medium Cost
howtobearealprogrammer icon

cc-mysql-mcp

by howtobearealprogrammer

Sec3

A Model Context Protocol (MCP) server enabling Claude Code to directly interact with local MySQL or MariaDB databases, primarily designed for Windows environments.

Setup Requirements

  • ⚠️Requires a local MySQL 5.7+ or MariaDB instance to connect.
  • ⚠️No built-in SQL query validation or safety measures; users must ensure the configured database user has restricted permissions to prevent destructive operations.
  • ⚠️Requires manual Node.js project setup (clone, `npm install`, `npm run build`) before running.
Review RequiredView Analysis
The server directly executes arbitrary SQL queries provided via the `execute_query` tool without any internal sanitization or validation. This poses a significant SQL injection vulnerability if the input is not strictly controlled by a trusted source. The `mysql2` library's `connection.query` is called with a raw query string, allowing for any SQL statement (including destructive ones like `DROP TABLE` or `DELETE FROM`) to be executed. The README explicitly warns about 'No safety rails' and recommends using a restricted database user, which is crucial but does not mitigate the code's inherent vulnerability to malicious or erroneous AI-generated SQL. No hardcoded secrets were found; configuration uses environment variables.
Updated: 2025-12-14GitHub→
0
0
Medium Cost
Shounak-Ghosh icon

fin-mcp

by Shounak-Ghosh

Sec8

Provides an AI-powered multi-agent system to analyze SEC 10-K filings for strategic risks, management tone, and financial summaries, exposed via an MCP server.

Setup Requirements

  • ⚠️Requires OpenAI API Key (Paid Service)
  • ⚠️Requires SEC API Key (as specified in README, though direct use in sec_tools.py is for user-agent)
  • ⚠️Python 3.11 or higher
Verified SafeView Analysis
The project uses standard libraries and practices for fetching public data and interacting with LLMs. No 'eval' or obvious malicious patterns found. API keys are handled via environment variables, which is good practice. The primary security considerations would be prompt injection for the LLM agents and proper access control if the MCP server is exposed publicly, which is a general concern for any API. The current setup focuses on local execution for the demo.
Updated: 2025-12-05GitHub→
PreviousPage 544 of 760Next