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
Activi-AI icon

The-Brain

by Activi-AI

Sec4

A server providing a memory store and audit log for AI agents like Claude Desktop and Claude Code, designed to store and search company decisions, notes, and code snippets.

Setup Requirements

  • ⚠️Requires Python 3.12+ for `better-sqlite3` native module compilation during `npm install`.
  • ⚠️The SQLite database path (`SQLITE_PATH`) must be writable; defaults to `./data/brain.sqlite`.
  • ⚠️For multi-tenant usage with the HTTP API, an external authentication and authorization layer (e.g., JWT validation, API Gateway) is CRITICAL to enforce `x-org-id`, `x-user-id`, `x-user-role`, and `x-project-id` header integrity and prevent spoofing. Without it, the system is fundamentally insecure for shared environments.
Review RequiredView Analysis
The HTTP API relies on `x-org-id`, `x-user-id`, `x-user-role`, and `x-project-id` headers for multi-tenancy context. If these headers are missing, hardcoded default values (`org-001`, `user-admin-001`, `admin`, `proj-001`) are used. This header-based authentication/authorization is highly susceptible to spoofing if the API is exposed to untrusted networks, allowing unauthorized access to data across different tenants or users. The `requireAdmin` function similarly relies solely on a header value. While prepared statements prevent SQL injection, the lack of robust authentication/authorization for multi-tenancy makes it unsafe for public deployment without an external security layer. The STDIO server uses fixed default context, acceptable for local desktop use but not for exposed scenarios. CORS is enabled globally without specific origin restrictions, posing a risk if deployed publicly. Error messages could potentially leak internal details.
Updated: 2026-01-18GitHub→
0
0
Low Cost
siddhant-ew icon

mcp-server-expenses

by siddhant-ew

Sec9

Manages and tracks personal expenses, storing them in a SQLite database via a FastMCP server.

Setup Requirements

  • ⚠️DB_PATH hardcoded to a Windows-specific path (C:\Projects\Personal\mcp-server-expenses), requiring modification for other environments.
  • ⚠️Requires Python 3.11 or higher.
  • ⚠️Requires 'fastmcp' library to be installed.
Verified SafeView Analysis
SQL queries are parameterized, mitigating SQL injection risks. No obvious hardcoded secrets or malicious patterns were found. The 'get_square' function in 'remote-mcp.py' calculates 'a**a', which could lead to very large numbers, potentially causing performance issues for extremely large inputs, but is not a direct security vulnerability like RCE. The DB_PATH is hardcoded to a Windows-specific path in 'main.py' and 'expense_tracker.py', which is a portability issue but not a security risk.
Updated: 2025-11-29GitHub→
0
0
Low Cost

mcp-demo-server

by MomentaryChen

Sec9

A demo server demonstrating basic real-time communication using the @mcp-socket/server library, sending welcome and 'Ping!' messages to connected clients.

Setup Requirements

  • ⚠️Requires Node.js and npm/yarn
  • ⚠️TypeScript project requires compilation (npm run build) before running the start command.
Verified SafeView Analysis
The code is simple and demonstrates basic server functionality without complex user input processing, 'eval' usage, or hardcoded secrets. It listens on port 8080 as expected for a server and primarily logs messages and sends fixed strings.
Updated: 2025-11-27GitHub→
0
0
Low Cost
dkasaj icon
Sec8

Enable AI assistants to play YouTube videos in the Tubist macOS app using natural language.

Setup Requirements

  • ⚠️Requires macOS 13 (Ventura) or above.
  • ⚠️Requires Tubist v1.3+ app to be installed.
  • ⚠️Requires manual configuration of Claude Desktop (or other MCP client) JSON file with correct absolute paths for Node.js and the server's `index.js`.
Verified SafeView Analysis
The server uses `child_process.spawn('open', [tubistURL])` to open URLs. The `tubistURL` is constructed using a YouTube URL provided by the AI assistant. While `open` is a trusted macOS command for delegating URL handling, the overall security relies on the Tubist application's robust handling of its custom URL scheme (`tubist://play?`). The server itself does not contain direct `eval` calls, unsanitized shell interpolations for arbitrary commands, or hardcoded secrets.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
yummysuperapp icon

bi-cloud-run-mcp-servers

by yummysuperapp

Sec9

Provides AI assistants (like Claude) with access to dbt models, metrics, and BigQuery data via an MCP server deployed on Google Cloud Run.

Setup Requirements

  • ⚠️Requires a Google Cloud Project with billing enabled.
  • ⚠️Requires a dbt Cloud account and API token.
  • ⚠️Manual update of Dockerfile to specify your dbt repository URL is required.
Verified SafeView Analysis
The project demonstrates strong security practices, emphasizing avoiding hardcoded credentials by using environment variables, GitHub Secrets, and Cloud Run Secret Manager. It provides clear documentation and setup guides for secure configuration, including proper `.gitignore` usage and separate service accounts for least privilege. The default `--allow-unauthenticated` for Cloud Run requires user attention for production hardening, but instructions to disable it are provided.
Updated: 2025-12-14GitHub→
0
0
Low Cost
tushargandhi77 icon
Sec9

Manages personal expenses by allowing users to add, list, and summarize expense entries through an API.

Setup Requirements

  • ⚠️Data is not persistent: The SQLite database is created in a temporary directory and will be lost if the server or environment restarts or the temporary directory is cleaned.
  • ⚠️Requires Python 3.11 or higher.
Verified SafeView Analysis
The code uses parameterized SQL queries, preventing SQL injection. No `eval` or similar dangerous functions are present, and no hardcoded secrets are identified. The database is initialized in a temporary directory, which isolates data but implies non-persistence across restarts if not explicitly handled.
Updated: 2025-12-03GitHub→
0
0
Medium Cost
zaevlad icon

HornetMCP

by zaevlad

Sec9

Provides a standalone Model Context Protocol (MCP) client to integrate smart contract vulnerability search directly into AI chat platforms like Claude Desktop.

Setup Requirements

  • ⚠️Requires an API key from hornetmcp.com, which may involve usage quotas and different tiers (free/paid).
  • ⚠️Requires Python 3.11 or higher.
  • ⚠️Requires an MCP-compatible AI client (e.g., Claude Desktop, Claude CLI, VS Code with Claude extension) for integration.
Verified SafeView Analysis
The client's source code does not contain 'eval', obfuscation, hardcoded secrets, or obvious malicious patterns. API keys are managed through `.env` files or environment variables, which is a secure practice, and `.gitignore` prevents accidental commits. The primary security consideration for users is that the client sends user-provided Solidity code snippets or natural language descriptions to a remote third-party API (`https://hornetmcp.com`) for analysis. Users should be aware of this data sharing and ensure they trust the remote service with their code.
Updated: 2026-01-12GitHub→
0
0
Medium Cost
abiheiri icon
Sec9

Provides tools for interacting with the pCloud API for cloud storage management, including listing, downloading, uploading, renaming, moving, deleting files and folders, and managing trash.

Setup Requirements

  • ⚠️Requires a pCloud account with valid `PCLOUD_USERNAME` and `PCLOUD_PASSWORD` set as environment variables.
  • ⚠️Dependencies must be installed using `uv` (recommended in README) or `pip` within a Python 3.11+ environment.
  • ⚠️Docker deployment requires building the `pcloud-mcp:1.0` Docker image first.
Verified SafeView Analysis
The server uses environment variables for sensitive credentials (`PCLOUD_USERNAME`, `PCLOUD_PASSWORD`), which is a good practice. It utilizes `httpx` for network communication and `pathlib` for robust file path handling, mitigating common injection and path traversal risks. Asynchronous tasks are managed with `asyncio`, and logging is directed to `stderr` to avoid interfering with MCP's `stdout` JSON-RPC communication. No direct use of dangerous functions like `eval` or `os.system` was observed. The authentication mechanism handles token generation and refresh, and API regions are configurable. Overall, the security posture appears robust for its intended function.
Updated: 2025-12-08GitHub→
0
0
Medium Cost

ssh-mcp-streamable

by Xxx00xxX33

Sec3

Establishes a multi-channel, streamable communication server over SSH, enabling secure and multiplexed data transfer.

Setup Requirements

  • ⚠️Requires SSH host key pair generation and configuration.
  • ⚠️Requires a specific network port (e.g., 22, 2222) to be open and accessible.
  • ⚠️Likely depends on specific Node.js modules for SSH handling (e.g., 'ssh2').
Review RequiredView Analysis
Without the actual source code, a definitive security audit is impossible. However, the project name 'ssh-mcp-streamable' indicates it likely implements or wraps an SSH server. SSH servers, by their nature, handle external, untrusted input and often involve executing commands or managing file descriptors. This makes them inherently high-risk components if not implemented with extreme care, requiring robust input validation and sanitization. Common vulnerabilities include command injection, improper authentication/authorization, denial-of-service, and mishandling of data streams. The absence of a README further increases the risk as no best practices or security considerations are documented. The potential for 'eval' or similar dangerous patterns is high in custom server implementations if not carefully managed. Given these inherent risks and the lack of inspectable code, a low security score is assigned.
Updated: 2025-11-25GitHub→
0
0
Medium Cost
Varnan-Tech icon

Meta-Ads-MCP_PixelPay

by Varnan-Tech

Sec9

Enables AI assistants to interact with the Meta Ads (Facebook/Instagram) Marketing API programmatically through natural language conversations for ad account, campaign, ad set, and ad management, analytics, and targeting.

Setup Requirements

  • ⚠️Requires manual setup of a Facebook App with Marketing API access, including configuring OAuth redirect URIs and obtaining App ID/Secret.
  • ⚠️Specific Facebook App permissions are needed (e.g., `ads_management`, `business_management`, `read_insights`), which may require App Review for broader use cases or specific roles for internal testing.
  • ⚠️Requires Python 3.10 or higher.
Verified SafeView Analysis
The server operates locally, significantly reducing external attack surface. It employs encryption (Fernet) for storing access tokens in a local SQLite database, and the README clearly advises setting a strong `TOKEN_ENCRYPTION_KEY`. Input validation is present for API parameters (e.g., object IDs, campaign data). OAuth token revocation on Meta's servers is explicitly implemented. The CORS configuration uses `allow_origins='*'` in development, which is a minor risk but is properly conditioned on `is_production` for production deployments. No obvious 'eval', 'exec', or 'os.system' calls are used with untrusted input.
Updated: 2025-11-24GitHub→
0
0
High Cost
Pond500 icon

rag-mcp-server

by Pond500

Sec1

This server implements a Multi-Knowledge Base RAG system, allowing AI agents to upload, manage, and semantically search documents across multiple knowledge bases through a FastAPI-based MCP (Model Context Protocol) API.

Setup Requirements

  • ⚠️Requires Docker for the Qdrant vector database.
  • ⚠️Requires an OpenAI-compatible LLM endpoint (e.g., Tokenmind) with an API key, which incurs usage costs.
  • ⚠️Hardcodes a dependency on a specific external 'Typhoon OCR API' (`http://3.113.24.61/...`) for PDF processing, which must be accessible and may incur separate costs.
  • ⚠️Requires Python 3.10+.
Review RequiredView Analysis
CRITICAL: The `LLM_API_KEY` in `app/config.py` has a hardcoded default value, which is a severe security vulnerability as it can lead to API key leakage if the `.env` file is not properly configured or if the server is run in an exposed environment. The `OCR_API_ENDPOINT` is hardcoded to an external IP, creating a single point of failure and potential data privacy concerns. The server is configured to listen on `0.0.0.0`, making it accessible from any network interface without any explicit authentication or authorization layer at the API level, relying solely on the integrating AI agent (e.g., Dify) for access control. This makes it highly vulnerable if exposed directly to the internet.
Updated: 2025-11-27GitHub→
0
0
Medium Cost
lianekai icon

mcp-dm8-server

by lianekai

Sec92

This server provides Model Context Protocol (MCP) tools for interacting with a DM8 database, enabling client applications (like AI agents) to browse table schemas and execute read-only SQL queries.

Setup Requirements

  • ⚠️Requires Node.js 18+ to be run with `--openssl-legacy-provider` due to the DM8 database driver's dependency on older OpenSSL algorithms.
  • ⚠️Requires a running DM8 database instance and a user account with appropriate permissions (read-only recommended).
  • ⚠️The 'dmdb' Node.js native driver may require specific system libraries to be installed as per DM8 official documentation.
Verified SafeView Analysis
The server includes comprehensive security fixes (version 1.1.0) addressing critical vulnerabilities found in prior versions. Key mitigations include: 1. Enhanced SQL injection protection with parameterization, robust identifier normalization, multi-statement detection, and blacklisting of dangerous functions/keywords/patterns (e.g., file operations, subquery writes, NULL bytes, excessive nesting). 2. Implementation of a connection pool to prevent resource exhaustion and improve performance. 3. Addition of query timeouts to prevent long-running queries. 4. Integration of a token-bucket based rate limiting mechanism. 5. Structured logging with sensitive data redaction. 6. Input length limits for queries and identifiers. The project explicitly states a security score of 92/100 post-fixes and is 'recommended for production use'.
Updated: 2025-11-27GitHub→
PreviousPage 350 of 760Next