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)

32
2
Low Cost
AerionDyseti icon

mcp-memory-server

by AerionDyseti

Sec6

Provides semantic memory storage for AI assistants, enabling them to store and retrieve decisions, patterns, and context across sessions using vector embeddings.

Setup Requirements

  • ⚠️Requires Bun 1.0+ runtime to be installed.
  • ⚠️Initial installation triggers a one-time download of a ~90MB ML model for embeddings, which can take a minute.
  • ⚠️Relies on native Node.js dependencies (`onnxruntime-node`, `sharp`) which can occasionally lead to platform-specific installation issues.
  • ⚠️Requires manual configuration in the MCP client's settings (e.g., `~/.claude/settings.json`).
Verified SafeView Analysis
The server operates locally by default (127.0.0.1:3271) reducing external network exposure. It uses `randomUUID` for internal ID generation, which is good. However, several database queries in `src/db/memory.repository.ts` construct `where` clauses using direct string interpolation for user-provided IDs (e.g., `id = '${id}'`). While IDs are typically UUIDs generated by trusted clients or internally, a malicious MCP client could send specially crafted strings (e.g., containing single quotes) as part of tool call arguments (`ids` in `delete_memories`, `get_memories`, `update_memories`), potentially leading to a form of SQL injection in the LanceDB queries. This is mitigated by its local-first, client-controlled nature, but remains a vulnerability pattern.
Updated: 2026-01-15GitHub→
32
2
Medium Cost
botextractai icon
Sec9

Builds a local MCP server with stock market tools for a LangChain AI agent to fetch financial data.

Setup Requirements

  • ⚠️Requires OpenAI API Key (Paid)
  • ⚠️Requires local `mcp_server.py` to be running in conjunction with the client
  • ⚠️Python dependencies need to be installed via pip
Verified SafeView Analysis
The server uses `stdio` for local communication, reducing network risks in this example. It relies on the security of `yfinance` and `fastmcp` libraries. `OPENAI_API_KEY` is loaded from `.env` file, which is good practice. No `eval` or direct shell command injection points from unsanitized user input were found. The regex for extracting tool messages is applied to internal `ToolMessage` string representation, not user input directly.
Updated: 2026-01-19GitHub→
32
2
Low Cost
Knuckles-Team icon

systems-manager

by Knuckles-Team

Sec1

This project provides a multi-OS system management CLI and an MCP server for automating system maintenance, application management, and configuration via AI or automation systems.

Setup Requirements

  • ⚠️Default Insecurity: The MCP server runs without authentication by default (AUTH_TYPE=none), which is a critical security risk. Users must explicitly configure authentication for secure operation.
  • ⚠️Operating System Specificity: Functionality relies on native package managers (apt, dnf, winget, etc.), requiring specific OS compatibility and underlying system tools.
  • ⚠️Elevated Privileges: Many core functionalities, including installation, updates, and feature management, require root or administrator permissions on the host system.
  • ⚠️Python 3.10+ and Specific Dependencies: Running natively requires Python 3.10+ and manual installation of packages like fastmcp, psutil, distro, requests and potentially eunomia-mcp if authorization is enabled.
Review RequiredView Analysis
The server exposes highly privileged 'run_command' and 'text_editor' tools. The 'run_command' tool can execute arbitrary commands on the host system, potentially with elevated (root/administrator) privileges and with 'shell=True', making it vulnerable to command injection if input is not carefully controlled, or allowing a malicious actor to run any command. The 'text_editor' tool allows arbitrary file creation, viewing, and modification, which can lead to system compromise or data loss. The default authentication type for the MCP server ('AUTH_TYPE=none' in docker-compose and CLI args) is highly insecure, allowing any client to invoke these powerful and destructive tools without authentication. While advanced authentication (JWT, OIDC) and authorization (Eunomia) are supported, they are not enabled by default, making initial deployments extremely risky.
Updated: 2026-01-19GitHub→
32
2
Low Cost
InfraMCP icon
Sec7

Enables AI agents to interact with Atlassian Cloud services (Jira, Confluence, Jira Service Management) for work documentation, issue management, and context understanding through a unified API.

Setup Requirements

  • ⚠️Requires an Atlassian OAuth 2.0 app to be created and configured in the Atlassian Developer Console.
  • ⚠️Specific OAuth scopes (e.g., `read:jira-work`, `write:page:confluence`, `read:servicedesk-request`, `offline_access`) must be explicitly configured in the Atlassian OAuth app's permissions.
  • ⚠️The OAuth callback URL must be set exactly to `http://localhost:8080/callback` in the Atlassian OAuth app settings, which can cause port conflicts locally.
Verified SafeView Analysis
The server implements OAuth 2.0 with PKCE for strong authentication. However, several security risks were identified in the project's own `SECURITY_ASSESSMENT.md` and through code analysis: 1. **Hardcoded Localhost Callback Server**: The OAuth callback server binds to `localhost:8080` without configuration, posing risks of port conflicts and potential local hijacking. 2. **Client Secret in Environment Variables**: Storing the OAuth client secret in plain text environment variables makes it visible in process lists and logs. 3. **Credential File Storage**: Access and refresh tokens are stored in a local JSON file (`~/.atlassian_mcp_credentials.json`) with `0600` permissions, but without encryption at rest, allowing access by anyone with access to the user's account and potentially persisting indefinitely. 4. **Limited HTTP Callback Server Security**: The temporary HTTP server for OAuth callbacks has minimal security, lacking comprehensive request validation and robust timeouts, making it potentially vulnerable to manipulation or DoS during the callback window. 5. **Error Information Disclosure**: Detailed error messages could potentially leak sensitive system information in a production environment. While the code itself does not appear to contain malicious patterns like `eval` or obfuscation, the handling of sensitive credentials locally and the network configuration for the OAuth callback require careful consideration for production deployments.
Updated: 2025-12-14GitHub→
32
2
Medium Cost
boecht icon

birre

by boecht

Sec9

BiRRe is a Model Context Protocol (MCP) server that translates a BitSight subscription into LLM-friendly tools for security rating retrieval and risk management.

Setup Requirements

  • ⚠️Requires a BitSight API key (implies a paid BitSight subscription).
  • ⚠️Requires Python 3.13+ runtime environment.
  • ⚠️Requires HTTPS network access to `api.bitsighttech.com`; custom CA certificates might be needed for corporate proxies.
Verified SafeView Analysis
The project follows good security practices. API keys are required as environment variables, not hardcoded. Network communication via `httpx` enforces TLS 1.2+ by default, with explicit options for custom CA bundles or disabling verification (with warnings) for testing. Input validation is performed for CLI arguments and some API parameters. No `eval` or `exec` calls were found. File operations are standard for configuration. The `--allow-insecure-tls` flag exists for diagnostics, which is a potential risk if misused, but is clearly documented and warns the user.
Updated: 2026-01-19GitHub→
32
2
Low Cost
Sec6

A Minecraft proxy server designed to integrate with Reddit, likely for features such as real-time notifications, bridging chat, or managing server events through Reddit.

Setup Requirements

  • ⚠️Requires Reddit API credentials (e.g., Client ID, Client Secret) to establish connection and interact with the Reddit platform.
  • ⚠️Requires a Node.js runtime environment to execute the server application.
  • ⚠️Likely requires configuration parameters for connecting to or acting as a proxy for a Minecraft server instance.
Review RequiredView Analysis
The contents of `index.js` and `package.json` were not provided, which prevents a detailed security audit for common risks like 'eval' usage, code obfuscation, or specific network vulnerabilities. General risks associated with any network-facing server and integration with external APIs (Reddit) apply. Without code inspection, specific guarantees cannot be made.
Updated: 2025-11-18GitHub→
32
91
Medium Cost
mahdin75 icon

gis-mcp

by mahdin75

Sec2

A Model Context Protocol (MCP) server that provides AI agents and LLMs with comprehensive GIS capabilities, enabling geospatial analysis, data gathering, and transformations through natural language.

Setup Requirements

  • ⚠️Requires Python 3.10 or higher.
  • ⚠️Specific features (e.g., visualization, climate data) require installation of optional dependencies (e.g., `uv pip install gis-mcp[climate]`).
  • ⚠️Climate data download (`cdsapi`) requires an external Copernicus Climate Data Store (CDS) account, API key, and acceptance of license agreements, stored in a `~/.cdsapirc` file.
  • ⚠️Critical security vulnerability related to arbitrary file access if user-provided paths are not strictly validated and confined.
Review RequiredView Analysis
CRITICAL: The `resolve_path` function in `src/gis_mcp/storage_config.py` allows user-provided absolute paths for file operations (upload, download, list, and data-specific output paths). This enables arbitrary file read/write/list operations anywhere on the server's filesystem, limited only by the permissions of the running process, leading to a severe path traversal / arbitrary file operation vulnerability. Additionally, when running in HTTP mode, the server defaults to `0.0.0.0` (all interfaces), making it publicly accessible.
Updated: 2025-12-21GitHub→
32
7
Medium Cost
StacklokLabs icon

plotting-mcp

by StacklokLabs

Sec8

A MCP server that transforms CSV data into beautiful visualizations for AI assistants and chat applications.

Setup Requirements

  • ⚠️Python 3.13+ required.
  • ⚠️Requires a writable filesystem for Matplotlib/Cartopy caching (especially in containerized environments).
  • ⚠️Docker is required for ToolHive deployment.
Verified SafeView Analysis
The server uses `json.loads` on user-provided `json_kwargs`. While Python's `json` module is generally safe against arbitrary code execution, extensive or malformed inputs could potentially lead to resource exhaustion or unexpected plotting library behavior. The server listens on `0.0.0.0`, requiring external network security. No `eval` or hardcoded secrets were found. The project includes a `SECURITY.md` and a responsible disclosure process.
Updated: 2026-01-07GitHub→
32
2
High Cost
ianaleck icon
Sec9

Provides a Model Context Protocol (MCP) server for seamless integration with the Harvest time tracking API, enabling AI clients to manage time entries, projects, clients, users, expenses, invoices, estimates, and reports.

Setup Requirements

  • ⚠️Requires a Harvest account with API access (Personal Access Token and Account ID).
  • ⚠️Requires manual configuration in an MCP-compatible client (e.g., Claude Desktop) with specific JSON syntax for `command` and `env` variables.
  • ⚠️Harvest API responses, especially for list operations with high `per_page` values, can be very verbose, potentially leading to high LLM token costs when the output is consumed by an AI.
Verified SafeView Analysis
The server demonstrates strong security practices. It uses Zod for comprehensive input validation on all tool arguments, significantly reducing the risk of injection attacks or malformed requests. Sensitive API credentials (Harvest Access Token and Account ID) are strictly loaded from environment variables, preventing hardcoding. Error handling is well-structured, wrapping exceptions in MCP-compliant error responses to prevent sensitive information disclosure. Dependencies like Axios, Express, Winston, and Zod are reputable. An optional API key authentication is available for the HTTP transport, which is a good security layer, although not mandatory by default for HTTP.
Updated: 2025-12-01GitHub→
32
2
Medium Cost
QuentinCody icon

entrez-mcp-server

by QuentinCody

Sec9

Provides a comprehensive Model Context Protocol (MCP) server that enables LLMs and applications to interact with NCBI biomedical databases (E-utilities, PubChem, PMC, BLAST) using language-native SDKs and advanced SQL-based data staging for analysis.

Setup Requirements

  • ⚠️An optional NCBI_API_KEY environment variable is highly recommended to increase API rate limits from 3 to 10 requests/second (3.3x faster).
  • ⚠️Designed for deployment on Cloudflare Workers, requiring a Cloudflare account for production use.
  • ⚠️For local development and SDK testing, the server needs to be run locally via 'npm start' or 'npx wrangler dev'.
Verified SafeView Analysis
The server explicitly enforces 'SELECT' only queries for staged data, mitigating SQL injection risks. NCBI API keys are loaded securely from environment variables, not hardcoded. External API calls are directed to well-known NCBI endpoints. No 'eval' or obfuscation detected. Hardcoded default email/tool are for identification with NCBI, not sensitive.
Updated: 2025-12-04GitHub→
32
2
Low Cost
jasonkneen icon

fluent-mcp

by jasonkneen

Sec9

A fluent API for building Model Context Protocol (MCP) servers and clients to orchestrate AI workflows and agent interactions across various platforms with built-in CRUD and transport management.

Setup Requirements

  • ⚠️Requires Node.js runtime (>=18.0.0).
  • ⚠️MCP is a protocol, so understanding its concepts (tools, resources, prompts) is necessary.
  • ⚠️Default resource storage is in-memory; persistent storage requires custom integration.
  • ⚠️Using SSE transport requires an Express.js server setup to handle HTTP requests.
Verified SafeView Analysis
The core library is built on `@modelcontextprotocol/sdk` and uses Zod for schema validation in tool definitions, which is good for input sanitization. The framework allows developers to define custom tool handlers, and the overall security depends on how these handlers are implemented by the user. No direct `eval` or explicit command injection vulnerabilities were found in the library's core implementation. Console output redirection for stdio transport is a good practice for MCP server integrity. Running the server with network transports (like SSE) on `0.0.0.0` is standard but requires proper environment setup (e.g., firewall) for production.
Updated: 2025-11-26GitHub→
32
2
Medium Cost
cvenkatreddy icon

playwright-mcp

by cvenkatreddy

Sec8

Provides a guide and example client project for generating and running Playwright UI and API tests using an MCP Playwright server with LLM agents.

Setup Requirements

  • ⚠️Requires Node.js v18 or newer.
  • ⚠️Requires installation of the '@playwright/mcp' package.
  • ⚠️Requires a compatible MCP client (e.g., Claude Desktop, Windsurf, VS Code) and integration with an LLM agent like Claude Sonnet 4 for test generation.
Verified SafeView Analysis
The provided source code for the client project, including test files and utility scripts, does not contain direct security vulnerabilities like 'eval' or hardcoded secrets. It relies on `npx @playwright/mcp@latest` to start the server, an external dependency from a trusted source (Microsoft). The primary security consideration arises from the nature of LLM-generated code, as the README states that generated test suites can vary. Running unreviewed, dynamically generated tests could pose risks if the LLM produces malicious code, although the provided example run commands explicitly target specific, benign test files included in the repository.
Updated: 2026-01-09GitHub→
PreviousPage 180 of 760Next