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
benhalverson icon

fusion-mcp

by benhalverson

Sec8

An MCP server for Autodesk Fusion 360, enabling data management, export, design automation, and webhooks interactions via a standardized protocol, deployed on Cloudflare Workers.

Setup Requirements

  • ⚠️Requires an Autodesk Forge/APS account with API credentials (usage beyond free tiers may incur costs).
  • ⚠️Requires a Cloudflare Workers account for deployment and Node.js 18+ for local development.
  • ⚠️Design Automation activities and webhook configurations require prior setup in the Autodesk Forge portal.
Verified SafeView Analysis
Secrets (Autodesk Forge Client ID/Secret) are correctly handled via Cloudflare Workers secrets, as indicated in `wrangler.toml` and setup instructions, preventing hardcoding. The server code does not use dangerous dynamic execution patterns like 'eval'. The `/webhook` endpoint explicitly notes that signature verification and event processing are left to the implementer; while the server itself doesn't perform immediate malicious actions with unverified webhooks, relying on unverified webhooks in a broader system can introduce security risks.
Updated: 2026-01-13GitHub→
0
0
Medium Cost
ST0MP-EE5 icon

ai-hub

by ST0MP-EE5

Sec8

The AI Hub MCP server acts as a central gateway for AI assistants to access and manage various tools, skills, and configurations, including DigitalOcean infrastructure.

Setup Requirements

  • ⚠️A DigitalOcean account and `DO_API_TOKEN` are required for the DigitalOcean MCP to function, implying paid services.
  • ⚠️Node.js (version 18.0.0 or higher) and npm must be installed on the host system.
  • ⚠️The main AI Hub server and each local MCP (e.g., `mcps/digitalocean`) require a separate build step (`npm run build`) before running.
Verified SafeView Analysis
The server employs good security practices including API key hashing, granular permissions, and standard web security middleware (helmet, cors, rate-limiting). Local MCPs are executed in isolated child processes, and external MCP routing is currently a placeholder, which mitigates immediate risks related to third-party integrations. The CLI utilizes `execSync` for administrative tasks, which is acceptable for a privileged user but highlights the need for secure CLI access. All secrets are managed via environment variables and API keys are stored as hashes. Potential future risks may arise with the full implementation of external MCP routing if not carefully secured against SSRF and malicious tool responses.
Updated: 2026-01-19GitHub→
0
0
Low Cost
hidao80 icon

memory_mcp

by hidao80

Sec8

A local memory management server for desktop extensions to store and manage user preferences, interests, and current status.

Setup Requirements

  • ⚠️Requires Python 3.8 or higher.
  • ⚠️Requires Node.js for generating the .dxt extension file via 'npm run package'.
  • ⚠️Requires a user to select a 'Memory file (.json)' location when enabled in the Claude Desktop extension settings.
Verified SafeView Analysis
The server performs local file I/O for memory data and logs. The memory file path is user-configurable via the desktop extension, which should handle path sanitization. The server listens on a local port (default 58902) as expected for an MCP, but this is a potential vector if other local processes are untrusted. No 'eval' or other highly dangerous functions were found. No hardcoded secrets.
Updated: 2026-01-17GitHub→
0
0
Low Cost

mediamath-mcp-mock

by Hypermindz-AI

Sec7

Provides a mock server for the MediaMath campaign management API, designed for development and testing of agent frameworks and integrations.

Setup Requirements

  • ⚠️Requires manual configuration of `JWT_SECRET` in `.env.local`.
  • ⚠️Not suitable for production use; designed solely for development and testing environments.
  • ⚠️Authentication requires specific mock credentials (e.g., admin@acme.com / password123) for initial testing.
Verified SafeView Analysis
Designed as a development/testing mock server; explicitly states it's not production-ready due to simplified authentication (no password hashing, in-memory tokens) and client validation. It is safe to run for its intended purpose, but not for sensitive data in a production environment.
Updated: 2025-11-19GitHub→
0
0
Low Cost
hamzasaleem22 icon

mcp-server-repository

by hamzasaleem22

Sec1

Analysis of the server's use case cannot be determined as the source code was not provided.

Setup Requirements

  • ⚠️Source code was not provided, therefore a comprehensive analysis of setup requirements and friction points is impossible.
Review RequiredView Analysis
Security audit cannot be performed as the source code for the 'mcp-server-repository' was not provided in the prompt. A score of 1 is assigned because safety cannot be verified without code.
Updated: 2025-12-02GitHub→
0
0
Medium Cost
BACH-AI-Tools icon

bachstudio-pubmedmcp

by BACH-AI-Tools

Sec9

Provides a Micro-service Container Protocol (MCP) server for searching and fetching biomedical literature abstracts from the PubMed database.

Setup Requirements

  • ⚠️Requires 'uv' package installed globally and 'uvx' in PATH.
  • ⚠️Python 3.12+ only.
Verified SafeView Analysis
The server code (src/pubmedmcp/__main__.py) is well-structured, utilizing Pydantic for robust input validation and relying on a dedicated client library (pubmedclient) for external API interactions, minimizing direct network risk. No 'eval' or arbitrary code execution is present in the server's operational logic. The 'bump_version.py' script, while using 'subprocess.run', is an external development utility and not part of the deployed server.
Updated: 2025-12-04GitHub→
0
0
Medium Cost
RiDDiX icon
Sec8

Enables AI assistants (like Claude) to manage and interact with a MountainOS (MOS) system by providing an interface via the Model Context Protocol (MCP).

Setup Requirements

  • ⚠️Requires a running MountainOS (MOS) system with its API accessible.
  • ⚠️Requires a MOS API authentication token or username/password to be provided via environment variables (many tools are 'admin only').
  • ⚠️Requires Node.js version 18.0.0 or higher.
Verified SafeView Analysis
The server acts as a proxy for the MOS API. Many operations require admin privileges on MOS, implying that the provided authentication token grants administrative access to the underlying system. Security relies heavily on the robustness of the MOS API itself and the secure management of the 'MOS_API_TOKEN' environment variable. The API client does not explicitly enforce HTTPS for 'MOS_API_URL', which could be a risk if connecting to a non-local MOS server over an insecure channel. No 'eval' or obvious malicious patterns were found in the provided source code.
Updated: 2026-01-16GitHub→
0
0
High Cost
yyf icon

MCP2OSC

by yyf

Sec5

Facilitates parametric control of multimedia devices and creative applications (like MaxMSP) using natural language prompts via a Large Language Model (LLM) such as Claude.

Setup Requirements

  • ⚠️Requires Claude Desktop Application for LLM integration.
  • ⚠️Requires Node.js version 18.0.0 or higher.
  • ⚠️Default OSC host/ports (9500 send, 9501 receive) may conflict with existing applications; manual configuration via environment variables may be necessary. The service manager attempts to detect MaxMSP to mitigate conflicts.
  • ⚠️WebSocket functionality requires 'ws' npm package to be installed, which might need to be explicitly added if not present by default.
Verified SafeView Analysis
The system presents several security considerations: 1. **ReDoS Vulnerability:** The `handleGetReceivedMessages` function in `mcp-server.js` constructs a regular expression from user-provided `addressPattern` (`new RegExp(addressPattern.replace(/\*/g, '.*'))`). If an LLM or a malicious user (if the endpoint were exposed) provides a crafted regex pattern, it could lead to a Regular Expression Denial of Service (ReDoS) attack, causing the server to hang or crash. 2. **API Exposure Risks:** The `enhanced-dashboard-server.js` exposes several API endpoints (e.g., `/api/patterns/upload`, `/api/test-osc`). * `/api/patterns/upload` allows posting new OSC patterns. While validation is performed, a large or complex JSON payload could potentially lead to resource exhaustion or malformed data being stored if not handled robustly, impacting subsequent parsing. * `/api/test-osc` allows sending arbitrary OSC messages to a specified address and arguments. If the dashboard is accessible from an untrusted network, this could be exploited to send arbitrary UDP packets, potentially causing a denial of service or other malicious activity on internal networks. 3. **JSON Parsing from Network Input:** Both `websocket-osc-controller.js` and `enhanced-dashboard-server.js` parse JSON from incoming network messages (`JSON.parse(data.toString())` in WebSockets, `req.body` in Express). Malformed JSON could potentially crash the server or consume excessive resources if not adequately protected by robust error handling and input validation (which is present to some extent for patterns but could be extended). 4. **UDP Communication:** OSC relies on UDP, which is connectionless and stateless. This makes it susceptible to spoofing and message injection if the network is not trusted, though this is inherent to UDP and not specific to the application's implementation. 5. **File System Operations:** The system extensively uses `fs` operations for logging and pattern storage. While atomic writes are implemented in `mcp-server.js`, concurrent writes or extremely large log/pattern files could introduce performance bottlenecks or, in extreme cases, file corruption if error handling or resource limits are not fully robust. The `shared-storage.js` uses synchronous file operations which can block the event loop.
Updated: 2025-11-24GitHub→
0
0
Low Cost
Sec5

This repository provides an example for deploying a remote Model Context Protocol (MCP) server on Cloudflare Workers, designed to host AI tools without requiring authentication.

Setup Requirements

  • ⚠️Requires a Cloudflare account for deployment to Cloudflare Workers.
  • ⚠️The server is explicitly designed without authentication, making it unsuitable for production use without significant modifications to add security.
  • ⚠️Connecting local MCP clients (e.g., Claude Desktop) requires installing and running the `mcp-remote` proxy.
Review RequiredView Analysis
The server is explicitly designed to be 'authless', meaning it exposes a public API endpoint without any authentication. While suitable for demonstration purposes, deploying this server in a production environment without adding robust authentication mechanisms would pose significant security risks, allowing unauthorized access and potential abuse of hosted tools. No other obvious vulnerabilities like 'eval' or obfuscation were noted from the provided information.
Updated: 2025-11-17GitHub→
0
0
Low Cost
gwyer icon
Sec8

A local, privacy-preserving Retrieval-Augmented Generation (RAG) system that combines semantic and keyword search to answer questions from user-provided documents, with an MCP server API for seamless integration with Claude Desktop.

Setup Requirements

  • ⚠️Requires Ollama to be installed and running locally (`ollama serve`).
  • ⚠️Requires specific Ollama models (`llama3.1:latest` for LLM, `nomic-embed-text` for embeddings) to be pulled manually via `ollama pull <model-name>`.
  • ⚠️Minimum 8GB RAM (16GB recommended) for efficient operation, especially with large datasets and local LLMs.
  • ⚠️The MCP server mode specifically requires Claude Desktop to be installed and configured with absolute paths in `claude_desktop_config.json`.
Verified SafeView Analysis
The system operates entirely locally, mitigating network-based data exfiltration risks. Configuration is handled via `config.yaml`, avoiding hardcoded secrets in the codebase. User-provided queries are processed through LangChain's RAG chain and structured query engine. While the `StructuredQueryEngine` internally uses `df.query()` for structured data, which can be a vector for injection if arbitrary user input were passed directly, the exposed MCP tools (`count_by_field`, `filter_dataset`) validate inputs by field and value, rather than raw query strings, significantly reducing this risk in the MCP context. The LLM prompt explicitly instructs the model to use 'ONLY the provided context' and 'NEVER make up or infer information', which aims to reduce hallucination and potential prompt injection leading to unintended actions, though LLM-based injection remains a theoretical challenge for any RAG system.
Updated: 2025-12-14GitHub→
0
0
Low Cost
bashithj icon
Sec8

This project demonstrates an AI agent using LangGraph and LangChain's MCP adapters to orchestrate calls to multiple local MCP services (Math and Weather tools) for processing user queries.

Setup Requirements

  • ⚠️The GROQ_API_KEY environment variable is required for the client application to run.
  • ⚠️The weather server (`weather.py`) must be running independently (e.g., by executing `python weather.py` in a separate terminal) before the client (`client.py`) can successfully connect to it.
  • ⚠️The project requires Python 3.13 or higher as specified in `pyproject.toml`.
Verified SafeView Analysis
API keys are loaded from environment variables, which is good practice. No 'eval', 'exec', or other immediately dangerous patterns were found in the provided source code. The client launches a local subprocess for the math server, which is generally acceptable for local development but pathing should be robustly controlled in production environments.
Updated: 2025-11-27GitHub→
0
0
Medium Cost
i9wa4 icon
Sec9

Provides an MCP (Multi-Cloud Platform) server to execute SQL queries and explore Unity Catalog metadata on Databricks.

Setup Requirements

  • ⚠️Requires 'uv' package manager for execution.
  • ⚠️Databricks authentication (host and token or client_id/client_secret) must be configured in ~/.databrickscfg or via environment variables.
  • ⚠️A Databricks SQL Warehouse ID is required for SQL execution and lineage features, configurable in ~/.databrickscfg or DATABRICKS_SQL_WAREHOUSE_ID environment variable.
  • ⚠️Python 3.11 or higher is required.
Verified SafeView Analysis
The server explicitly blocks a comprehensive list of potentially dangerous SQL DDL/DML commands (DROP, DELETE, TRUNCATE, ALTER, CREATE, INSERT, UPDATE, MERGE, GRANT, REVOKE) to prevent accidental or malicious destructive operations. Authentication relies on standard Databricks SDK mechanisms (environment variables or ~/.databrickscfg), which are secure if credentials are managed properly. No 'eval' or other dynamic code execution found. No hardcoded secrets.
Updated: 2026-01-19GitHub→
PreviousPage 746 of 760Next