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
Medium Cost
steadybit icon

mcp

by steadybit

Sec8

The Steadybit MCP Server enables LLM tools to interact with the Steadybit chaos engineering platform for managing experiments, executions, and other Steadybit resources.

Setup Requirements

  • ⚠️Requires a Steadybit account and API token.
  • ⚠️The 'create_experiment_from_template' functionality must be explicitly enabled via environment variable (e.g., CAPABILITIES_ENABLED_0=CREATE_EXPERIMENT_FROM_TEMPLATE).
  • ⚠️Requires a Java Runtime Environment (if running locally from JAR) or Docker (if using the Docker image).
Verified SafeView Analysis
The server acts as a wrapper for the Steadybit API, making authenticated HTTP requests. It uses environment variables for sensitive API tokens, preventing hardcoding. Input validation for API URL and token is present. There are no obvious signs of 'eval' or direct shell command execution from user inputs. The primary risk is inherent to any API wrapper, relying on the security of the upstream Steadybit API and proper management of the API_TOKEN.
Updated: 2026-01-12GitHub→
0
0
Low Cost
gvdlga icon
Sec9

This MCP server provides an API wrapper for AssemblyAI, enabling audio transcription and management functionalities such as transcribing files, checking status, listing, and deleting transcriptions.

Setup Requirements

  • ⚠️Requires an AssemblyAI API Key (Paid service).
  • ⚠️Requires Node.js runtime environment.
  • ⚠️Requires running 'npm install' and 'npm run build' commands to prepare the application.
Verified SafeView Analysis
The server uses environment variables for API keys (`ASSEMBLY_API_KEY`) and relies on the `@geniusagents/mcp` library for potential session-based API key management, which is a good practice. No direct 'eval' or other obviously dangerous patterns were found. The server acts as a proxy, so the primary security consideration would be the secure management of the AssemblyAI API key and the `@geniusagents/mcp` framework's security.
Updated: 2025-11-20GitHub→
0
0
Medium Cost

cnxt-clipx-mcp

by thexdesk

Sec9

Generates highlight reels for sports and TV shows by processing data from sources like ESPN and curated TV registries, rendering them using FFmpeg, and optionally uploading to Google Cloud Storage.

Setup Requirements

  • ⚠️Requires Node.js and npm for setup and execution.
  • ⚠️Requires `ffmpeg` and `ffprobe` binaries to be installed and available in the system's PATH for the FFmpeg server to function.
  • ⚠️Requires `npm run sync:sports` to generate `sports-registry.json` for the sports server.
  • ⚠️Google Cloud credentials must be configured (e.g., via `GOOGLE_APPLICATION_CREDENTIALS` environment variable or Application Default Credentials) for GCS upload functionality in the FFmpeg server.
Verified SafeView Analysis
The server utilizes `child_process.spawn` for executing `ffmpeg` and `ffprobe` commands. Arguments are passed as an array, which is a safer practice against shell injection compared to using `shell: true` or `child_process.exec`. The Google Cloud Storage integration relies on standard authentication methods (`GOOGLE_APPLICATION_CREDENTIALS` or Application Default Credentials), reducing the risk of hardcoded secrets. External data fetching from ESPN is done via `fetch`. No obvious `eval` or malicious patterns were found. Relying on properly configured Google Cloud credentials is key for secure GCS operations.
Updated: 2025-11-27GitHub→
0
0
Medium Cost
parrisma icon

mcpnp

by parrisma

Sec9

Augment LLM math capabilities by exposing common numpy and scipy operations as a Micro-Capability Platform (MCP) server.

Setup Requirements

  • ⚠️Requires Python 3.x with numpy, scipy, fastmcp, and pydantic installed (or use Docker).
  • ⚠️Docker is highly recommended for building and running the server, with provided build scripts.
  • ⚠️The server defaults to binding on 0.0.0.0:9124, ensure this port is available and network access is controlled if exposed publicly.
Verified SafeView Analysis
The server uses FastMCP for tool exposure and Pydantic for robust input validation, significantly mitigating injection risks. It leverages standard and well-audited libraries (numpy, scipy). No 'eval' or 'exec' is used dynamically. Error handling is present to catch exceptions during numerical operations. The test client uses a placeholder bearer token, indicating potential for API authentication. The server binds to 0.0.0.0 by default, which is common for containerized services but exposes it on all interfaces, though this is configurable.
Updated: 2025-12-03GitHub→
0
0
High Cost
v3nom icon

toon-fetch

by v3nom

Sec5

Fetches web content from a given URL, cleans it, converts it to Markdown, and optionally processes it using a local language model to return structured data in TOON format for AI agents.

Setup Requirements

  • ⚠️Requires Chromium/browser dependencies for Puppeteer, which can be large downloads and may have system-specific installation requirements.
  • ⚠️Requires an initial download of the LaMini-Flan-T5-248M LLM model (~248MB) on first use, impacting startup time and disk space.
  • ⚠️Demands significant local compute resources (CPU, RAM) for running headless Chrome and local LLM inference per call, potentially making it unsuitable for low-resource environments or high-throughput usage without careful scaling.
Review RequiredView Analysis
The server uses Puppeteer to fetch content from arbitrary URLs. Running a headless browser to visit untrusted external websites can expose the host system to security risks (e.g., Server-Side Request Forgery if not properly isolated, or potential exploits if the browser itself is compromised). No explicit sandboxing mechanisms for Puppeteer are mentioned in the provided code. Additionally, the local LLM downloads its model weights on first use, requiring external network access and disk space, which could be a vector for supply chain attacks if the model source were compromised.
Updated: 2025-12-13GitHub→
0
0
Low Cost
cmalpass icon

mcp-presidio

by cmalpass

Sec9

Provides PII (Personally Identifiable Information) detection and anonymization capabilities using Microsoft Presidio, enabling LLMs to safely handle sensitive data.

Setup Requirements

  • ⚠️Requires Python 3.10 or higher.
  • ⚠️Requires downloading spaCy language models (e.g., en_core_web_lg) for optimal PII detection accuracy.
  • ⚠️Docker (20.10+) and Docker Compose (optional) are required for containerized deployment.
Verified SafeView Analysis
The server processes all data locally using `stdio` transport, eliminating network PII leakage from the server itself. It clearly warns users about the inherent risk of sending PII to LLM providers *before* the server processes it. No 'eval' or obvious hardcoded secrets are used for core functionality. User-provided regex for custom recognizers could theoretically be a vector for ReDoS if not robustly handled by the underlying Presidio library, but this is a standard pattern for such tools. Docker containers are configured to run as a non-root user.
Updated: 2025-12-10GitHub→
0
0
Medium Cost
thousandmiles icon

lsp-mcp-server

by thousandmiles

Sec8

Provides semantic code analysis capabilities, such as go-to-definition and find-references, by bridging Model Context Protocol (MCP) tool calls to a Language Server Protocol (LSP) server.

Setup Requirements

  • ⚠️Requires Node.js (v20+ due to `typescript-language-server` dependency requirements).
  • ⚠️Requires npm to install dependencies (`npm install`) and build the project (`npm run build`).
  • ⚠️The `typescript-language-server` needs to be successfully installed as a local dependency.
Verified SafeView Analysis
The server spawns a local `typescript-language-server` instance and reads local files via `fs.readFile` based on paths provided as tool arguments. While this is necessary for its functionality, it operates with the permissions of the user running it and can access any file the user has access to within the project root. An agent controlling the `filePath` or `query` arguments could potentially read or process sensitive local files, though no direct code injection or arbitrary command execution vulnerabilities (like `eval`) are present. The primary risk is the scope of file system access granted to the process and the data an agent might request from it.
Updated: 2025-11-22GitHub→
0
0
Medium Cost
duquesnay icon

miro-remote-mcp

by duquesnay

Sec9

Enables Claude AI to programmatically create and manipulate Miro boards for visualization and collaboration through natural language commands.

Setup Requirements

  • ⚠️Requires a Miro OAuth App to be created in Miro settings (Client ID, Client Secret, Redirect URI).
  • ⚠️Requires Node.js 16+.
  • ⚠️Initial OAuth2 tokens must be obtained by running `npm run oauth` and configured in the environment.
  • ⚠️OAuth access tokens expire hourly; the server automatically refreshes them, but if the refresh token also expires, manual reauthentication is required via the `/oauth/authorize` endpoint or `npm run oauth`.
Verified SafeView Analysis
The server explicitly handles sensitive data by reading OAuth credentials and tokens from environment variables or a specified file (`/data/tokens.json`). It uses base64 encoding for environment variables in some deployment contexts. There's a refresh lock to prevent race conditions during token refresh. All configuration files and token storage locations are documented as git-ignored. No 'eval' or direct malicious patterns were identified. The OAuth helper runs a temporary local server only for initial token acquisition. The overall design follows good security practices for a standalone application, but token files could be compromised if the hosting environment is not secure.
Updated: 2026-01-09GitHub→
0
0
Medium Cost
bajpainaman icon

DeltaMCP

by bajpainaman

Sec9

Enhances AI assistants with syntax-highlighted and colorized Git diffs, blame, and grep output via Model Context Protocol.

Setup Requirements

  • ⚠️Requires git-delta (installed via Cargo).
  • ⚠️Requires Git.
  • ⚠️Requires Python 3.10+.
  • ⚠️MCP client configuration requires absolute paths to the server directory.
  • ⚠️The client application must be fully restarted after configuration changes for the server to be recognized.
Verified SafeView Analysis
The server demonstrates strong security practices. Input validation is rigorously applied to file paths (preventing path traversal) and commit ranges (using regex patterns). Grep tools are whitelisted to prevent arbitrary command execution. Subprocess execution (`asyncio.create_subprocess_exec`) is used with command lists, not shell strings, mitigating command injection. The server operates over STDIO, minimizing network attack surface, and generated HTML files for browser viewing are local (`file://` URLs). No hardcoded secrets were found. Logging is directed to stderr to prevent interference with STDIO communication.
Updated: 2025-12-14GitHub→
0
0
Medium Cost
amberofficil icon

mcp_server_yt

by amberofficil

Sec1

Unable to determine the specific use case as no source code was provided for analysis.

Review RequiredView Analysis
CRITICAL: No source code was provided for analysis. Therefore, a thorough security audit for 'eval', obfuscation, network risks, hardcoded secrets, or malicious patterns could not be performed. Running this server without reviewing its code is extremely risky and is strongly advised against, as it may contain malicious patterns, hardcoded secrets, or unpatched vulnerabilities. The low score reflects this complete lack of visibility.
Updated: 2025-11-27GitHub→
0
0
Medium Cost
akbfar icon
Sec10

This application allows users to connect to a Zomato MCP Server via Claude Desktop to search for restaurants, browse menus, and place food orders using AI technologies.

Setup Requirements

  • ⚠️Requires downloading and installing an executable (.exe or .zip) from an external release page, which carries security risks without source code verification.
  • ⚠️Requires an internet connection for fetching restaurant data.
  • ⚠️Requires server details (API key and endpoint URL) from Zomato, to be entered in the application's settings.
Review RequiredView Analysis
The only 'source code' provided for analysis is the `Readme.md` file. This file itself contains no direct security risks, malicious patterns, `eval` statements, obfuscation, or hardcoded secrets. However, the `Readme.md` describes a desktop client application that requires users to download and install an executable (`.exe` or `.zip`) from a GitHub release page. A comprehensive security audit of the actual client application's binary logic or its source code (which was not provided) would be necessary to assess its real-world security posture. Running arbitrary executables without source code review carries inherent risks.
Updated: 2026-01-19GitHub→
0
0
Low Cost
mayorGonzalez icon

mcp-server-prueba

by mayorGonzalez

Sec9

Exposes VoltAgent AI agents and their integrated tools over the Model Context Protocol (MCP) using a Hono HTTP server, enabling discovery and invocation by compatible IDEs.

Setup Requirements

  • ⚠️Requires Node.js 20+ and pnpm.
  • ⚠️OPENAI_API_KEY must be configured in your environment for AI agent functionality.
  • ⚠️BUILDERBOT_API_KEY must be configured in your environment to interact with BuilderBot project management tools.
Verified SafeView Analysis
The server leverages established frameworks like VoltAgent, Hono, and Pino for logging, indicating good development practices. API keys (OPENAI_API_KEY, BUILDERBOT_API_KEY) are explicitly managed as environment variables, preventing hardcoding. Input validation is performed using Zod schemas for tools. Network exposure is primarily local (port 3141 by default) and via a well-defined protocol (MCP). The use of BuilderBotApiClient involves external API calls, which are handled with error checking. No obvious 'eval' or malicious code patterns were found in the provided snippets. The security score reflects confidence in the framework's design and a clear separation of concerns, while acknowledging that any networked application carries inherent risks.
Updated: 2025-11-30GitHub→
PreviousPage 679 of 713Next