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
Low Cost
madhaviai icon
Sec10

A Model Context Protocol (MCP) server that provides tools to retrieve professional profile information, skills, education, and contact details for an AI assistant or developer environment.

Setup Requirements

  • ⚠️Requires the 'mcp-server' Python package to be installed (e.g., `pip install mcp-server`).
  • ⚠️The 'run_command' provided in the `mcp.json` configuration uses an absolute path (`/Users/madhavik/...`) which will need to be adjusted by other users to match their local file system.
  • ⚠️Requires Python 3.x, likely 3.8+ due to asyncio usage.
Verified SafeView Analysis
The server operates by providing static, read-only professional profile data. It does not perform any file I/O, network requests, or execute arbitrary code based on user input. No 'eval', 'exec', 'os.system', or similar dangerous functions are used. All data is hardcoded within the script. Communication is via standard I/O, making it inherently secure for its intended use case.
Updated: 2025-12-12GitHub→
0
0
Medium Cost
ramailo1 icon
Sec4

DriftGuard is an MCP server that acts as a 'Traffic Controller' for AI agents, enforcing a strict state machine to prevent 'code drift' during development tasks.

Setup Requirements

  • ⚠️Requires Node.js version 18 or higher to run.
  • ⚠️Must be run within a Git repository to fully utilize L3 Git Notes (audit trails, risk analysis, integrity checks).
  • ⚠️The system grants the AI agent the ability to define and execute arbitrary shell commands (e.g., for `dg_verify`), necessitating trust in the agent or a sandboxed execution environment.
Verified SafeView Analysis
The server's core functionality involves executing arbitrary commands (`testCommand`) provided by the AI agent via the `dg_verify` tool. While this is an intended feature to allow agents to run tests, it means the server can be an arbitrary code execution vector if a malicious or compromised agent interacts with it, or if untrusted commands are passed to the agent. No explicit hardcoded secrets or direct `eval` calls were found. File system operations are scoped to `.driftguard/` for internal state and project files as claimed by tasks.
Updated: 2026-01-18GitHub→
0
0
Medium Cost
algtools icon

mcp-server

by algtools

Sec8

Deploys an authentication-less Model Context Protocol (MCP) server on Cloudflare Workers, providing AI-powered tools for mathematical calculations and RAG-based search.

Setup Requirements

  • ⚠️Requires a Cloudflare Workers account and project setup.
  • ⚠️Requires Workers AI (AI binding) to be configured in your Cloudflare Worker project.
  • ⚠️Requires a Cloudflare Durable Object Namespace binding (MCP_OBJECT) for state management.
  • ⚠️The 'searchCursorRules' tool requires an existing Cloudflare AI AutoRAG vector store, defaulting to 'algtools-cursor-rules-rag', or an ID provided via AUTORAG_ID environment variable.
Verified SafeView Analysis
The server is explicitly designed to be 'auth-less', meaning any client can access its tools. While input validation (zod) is used for tool arguments and external calls are directed to Cloudflare AI services, the inherent nature of 'auth-less' means it should not be used for sensitive operations or exposed without proper network controls if authentication is desired. There are no obvious code-level vulnerabilities like arbitrary code execution (eval) or hardcoded secrets beyond its intended functionality.
Updated: 2025-12-03GitHub→
0
0
Low Cost
scanzy icon
Sec1

This project provides an interface for AI to programmatically edit 2D CAD files in Draftsight on Windows using Win32 COM APIs.

Setup Requirements

  • ⚠️Requires Draftsight CAD software to be installed on a Windows machine.
  • ⚠️Specific Windows environment and permissions may be needed for Win32 COM API interaction.
Review RequiredView Analysis
The provided source code consists only of the README.md file. A security audit of the actual implementation is not possible without the relevant code files. Interacting with local applications via Win32 COM APIs carries inherent risks if not implemented securely, which cannot be assessed here.
Updated: 2025-11-25GitHub→
0
0
High Cost
uditsanghvi07 icon
Sec9

An agentic expense processing system for zero-shot data extraction from invoice images and autonomous expense classification into structured reports.

Setup Requirements

  • ⚠️Requires an external Multimodal LLM service (e.g., OpenAI, Google Gemini, Anthropic) for zero-shot data extraction and autonomous classification, which typically involves API keys and potential paid usage.
  • ⚠️Requires multiple Python processes to be run concurrently: the MCP server (main.py), a FastAPI backend (backend.py), and a Streamlit frontend (frontend.py).
Verified SafeView Analysis
The server uses SQLite with parameterized queries, mitigating SQL injection risks. No direct 'eval' or other dangerous functions are present. Database file paths are relative and localized. No hardcoded sensitive credentials were found. The use of 'check_same_thread=False' for SQLite in backend.py is common with FastAPI but should be understood for potential concurrency implications, though not a direct security flaw in this context.
Updated: 2025-11-20GitHub→
0
0
Medium Cost
ry-ops icon
Sec9

Manages and monitors Kubernetes (K3s) clusters through a Model Context Protocol (MCP) server, enabling AI-powered cluster management via Claude.

Setup Requirements

  • ⚠️Requires a valid Kubernetes `kubeconfig` file with appropriate RBAC permissions, and its path must be provided via the `KUBECONFIG` environment variable. Updating the `server` field in the `kubeconfig` to the correct K3s API IP/hostname is often necessary.
  • ⚠️Requires network connectivity from the machine running the MCP server to the K3s cluster's API server.
  • ⚠️Requires Python 3.10 or higher and the `uv` package manager for dependency management and execution.
Verified SafeView Analysis
The server leverages standard Kubernetes `kubeconfig` for authentication and authorization. Security critically depends on keeping the `kubeconfig` file secure, setting appropriate file permissions (e.g., 600), and configuring Kubernetes RBAC (Role-Based Access Control) to grant only necessary permissions. The code itself does not introduce hardcoded secrets or obvious vulnerabilities, and documentation explicitly highlights security best practices for `kubeconfig` and RBAC.
Updated: 2025-12-17GitHub→
0
0
High Cost
talianzakowski icon

py2to3mcp

by talianzakowski

Sec6

Facilitates Python 2 to 3 code migration, provides file system manipulation, and offers code search capabilities for AI-driven development.

Setup Requirements

  • ⚠️Requires Python 3.9 or higher.
  • ⚠️Requires specific configuration in `~/.claude.json` with absolute paths for each MCP server.
  • ⚠️The `py2to3-migration` server recommends installing `fissix` (`pip install fissix`) for full functionality and file conversion; otherwise, it falls back to the system's `2to3` command which may not be present or compatible.
Verified SafeView Analysis
The servers expose powerful file system and code execution capabilities. The `filesystem_server` allows reading and writing arbitrary files/directories (within process permissions and defined limits). The `py2to3_server` uses `subprocess.run` to execute external `2to3` or `fissix` commands for code transformation. While these are legitimate tools, the use of `subprocess` and broad file system access introduces a risk vector. A compromised or malicious calling agent could potentially misuse these tools to read/write sensitive files or execute arbitrary code on the host system. Internal `eval` is not present in server logic, only shown in a Python 2 example.
Updated: 2025-11-23GitHub→
0
0
Medium Cost
RheaBose icon

ilp-mcp-server

by RheaBose

Sec8

Enables Large Language Models (LLMs) to interact with a drone delivery system through natural language queries, planning deliveries, checking availability, and visualizing routes.

Setup Requirements

  • ⚠️Requires a separately running 'ILP CW2 Service' on http://localhost:8080 or specified by ILP_API_URL.
  • ⚠️Node.js 18+ is required.
  • ⚠️For Claude Desktop integration, `npm link` and specific configuration file editing are needed.
Verified SafeView Analysis
The server connects to a local ILP CW2 API (http://localhost:8080) by default, which limits external network attack surface. It uses `@modelcontextprotocol/sdk` and `zod` for tool schema validation, a good practice for input sanitization. No `eval` or obvious hardcoded secrets. The primary risk would stem from vulnerabilities in the underlying ILP CW2 service or if the `ILP_API_URL` environment variable is misconfigured to point to an untrusted external service.
Updated: 2025-12-01GitHub→
0
0
High Cost
Zehra-code743 icon

Quiz-App

by Zehra-code743

Sec9

Transforms academic PDFs into AI-generated structured study notes and interactive quizzes for learning and exam preparation.

Setup Requirements

  • ⚠️Requires GEMINI_API_KEY (Paid service)
  • ⚠️Python 3.11 or higher is required
Verified SafeView Analysis
The application handles API keys securely using environment variables or Streamlit secrets. It writes user-uploaded PDF files to a temporary directory, which is common practice but always carries a minor risk if not strictly managed, though `os.path.join` is used to mitigate path traversal. Text extraction from PDFs is done with `pypdf`, which is generally safe for its purpose. There are no direct `eval` or command injection vulnerabilities identified. AI model output should always be reviewed by the user.
Updated: 2025-11-24GitHub→
0
0
Low Cost
matiasbattocchia icon

google-mcp

by matiasbattocchia

Sec4

A public Model Context Protocol (MCP) server that enables AI assistants to access Google Calendar and Sheets without requiring users to set up their own Google Cloud project.

Setup Requirements

  • ⚠️Requires a Google Cloud project with Google Calendar and Sheets APIs enabled for self-hosting.
  • ⚠️Requires creating OAuth 2.0 Web application credentials in Google Cloud for self-hosting.
  • ⚠️Requires a Cloudflare account and D1 database setup for self-hosting.
Verified SafeView Analysis
The server stores Google OAuth refresh tokens and the generated API keys directly in the D1 database without application-level encryption or hashing. While Cloudflare D1 provides encryption at rest, a compromise of the database (e.g., through an SQL injection vulnerability or a Cloudflare internal breach) would expose all user API keys and refresh tokens. This contradicts the README's claims of 'Hashed' for API Keys and 'Encrypted' for OAuth Tokens, which are misleading if interpreted as application-level security measures. Refresh tokens grant long-term access to a user's Google account, making their direct storage a critical vulnerability. Other security aspects like input validation (Zod), proper OAuth flow implementation, and use of environment variables for secrets are well-handled. There is no usage of 'eval' or obvious malicious patterns.
Updated: 2026-01-16GitHub→
0
0
Low Cost
BACH-AI-Tools icon

NASA-MCP-server

by BACH-AI-Tools

Sec8

Provides a standardized Model Context Protocol (MCP) interface for AI models to interact with various NASA and JPL APIs.

Setup Requirements

  • ⚠️Requires a NASA API Key (available for free at api.nasa.gov).
  • ⚠️Requires Node.js (>=18 as specified by @modelcontextprotocol/sdk and package.json).
  • ⚠️Manual installation involves TypeScript compilation (`npm install` then `npm run build`).
Verified SafeView Analysis
The server uses `dotenv` for environment variable management and `zod` for robust input validation, which are good security practices. API keys are expected via environment variables or command-line arguments, preventing hardcoding. External API requests include timeouts. Dynamic module imports for handlers are used but appear to be constrained by a predefined set of tool names, reducing the risk of arbitrary code execution. Minor concerns exist regarding the direct manipulation of the global object for registering MCP tools, although the naming convention (`mcp__nasa...`) limits potential collisions. Overall, solid security hygiene for an API proxy.
Updated: 2025-12-04GitHub→
0
0
Medium Cost
MagedDevOps icon

mcp-server

by MagedDevOps

Sec1

Provides a Model Context Protocol (MCP) server for AlSalam Hospital management operations, including appointment booking, patient services, and WhatsApp integration.

Setup Requirements

  • ⚠️Requires Node.js version 22 or higher.
  • ⚠️Requires a valid WhatsApp Business API bearer token (which MUST replace the hardcoded one for secure operation).
  • ⚠️External API dependency on AlSalam Hospital's UAT API (https://salemuatapi.alsalamhosp.com:446).
Review RequiredView Analysis
CRITICAL: The WhatsApp API bearer token is hardcoded directly in the `send_whatsapp_message` tool's source code (`server.js`). This is a severe security vulnerability as it grants full access to send messages from the linked WhatsApp Business Account to anyone with access to the codebase. It should be stored as an environment variable and loaded securely. CORS is set to `Access-Control-Allow-Origin: *`, which is overly permissive. While the README mentions 'CORS support for browser compatibility', this 'allow all' setting is generally not recommended for APIs handling sensitive patient data unless strict IP-based access controls or other authentication mechanisms are in place upstream. The `REQUEST_TIMEOUT` is defined as an environment variable in the README but is hardcoded as a constant (`50000`ms) within `server.js`, meaning it cannot be configured via environment variables as advertised. Input validation is implemented using Zod schemas for tool arguments, which is a good practice.
Updated: 2025-12-18GitHub→
PreviousPage 401 of 713Next