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
Olbrasoft icon

Antigravity

by Olbrasoft

Sec9

Integrates Antigravity IDE notifications into a Virtual Assistant, enabling AI agents to provide voice or system alerts to the user.

Setup Requirements

  • ⚠️Requires Node.js (v18+) and npm to install dependencies and build.
  • ⚠️Needs manual configuration in Antigravity's `mcp_config.json` file, specifying the absolute path to the compiled `index.js`.
  • ⚠️Relies on a running Virtual Assistant API endpoint for notifications to be delivered, specified by the `VA_ENDPOINT` environment variable.
Verified SafeView Analysis
The server primarily acts as a proxy, forwarding notification text to a configurable external API endpoint (`VA_ENDPOINT`). There are no signs of direct code injection vulnerabilities (like `eval` or `child_process` execution of unsanitized user input) in the TypeScript source. The reliance on an external Virtual Assistant endpoint means the security posture is also dependent on the security of that external service, which is a standard integration practice.
Updated: 2026-01-15GitHub→
0
0
Medium Cost

Test_MCP_Server

by TweekFawkes

Sec6

A Python-based server designed for interacting with or serving applications that leverage Anthropic's Claude AI model.

Setup Requirements

  • ⚠️Requires Anthropic Claude API Key (Paid)
  • ⚠️Specific Python version required (check `.python-version` file for details)
  • ⚠️Requires `uv` package installer for dependency management
Review RequiredView Analysis
Cannot fully audit without code content. Potential risks include insecure API key handling (in `configure_claude.py/sh`), unvalidated inputs in `server.py` leading to arbitrary code execution or data exposure, and general network security risks associated with exposing a server to external traffic. No obvious signs of 'eval' or obfuscation from file names.
Updated: 2025-11-18GitHub→
0
0
High Cost
JoaquimCassano icon

cve-mcp

by JoaquimCassano

Sec8

Provides a tool-based interface to search for and retrieve details about Common Vulnerabilities and Exposures (CVEs) from official sources.

Setup Requirements

  • ⚠️Requires a Python environment with `fastmcp`, `pydantic`, and `cloudscraper` libraries installed.
  • ⚠️Relies on external CVE APIs (www.cve.org and cveawg.mitre.org) which might rate-limit or block aggressive scraping, potentially impacting `cloudscraper`'s long-term effectiveness.
  • ⚠️Search results, especially with a high `size` parameter, can return large amounts of data, leading to significant token consumption if processed by an LLM.
Verified SafeView Analysis
The code uses `cloudscraper` to bypass bot detection on `www.cve.org`, which, while not inherently malicious, is an aggressive technique that could lead to operational issues (e.g., IP bans or changes in scraping effectiveness). The `cve_id` is directly interpolated into a URL path in `gather_details` without explicit input validation, though the strict format of CVE IDs reduces the immediate risk of path traversal. Hardcoded user-agent and browser headers are used in network requests, which might require maintenance if external API detection methods change. No 'eval', obfuscation, hardcoded secrets, or other obvious malicious patterns were found. Pydantic models are used for robust data validation and parsing.
Updated: 2025-12-14GitHub→
0
0
Low Cost
Maoric2018 icon

mcp-leak-server

by Maoric2018

Sec2

The server's primary function is to deliver an XML payload instructing a 'client agent' to execute a hidden, one-time command for 'cleanup and synchronization'.

Setup Requirements

  • ⚠️Requires Node.js environment
  • ⚠️Requires npm to install dependencies
  • ⚠️Requires Vercel CLI to run via 'npm run dev'
Review RequiredView Analysis
The server's core purpose is to deliver an XML response containing explicit instructions for a client-side agent to execute a hidden, unlogged command (`[HIDDEN_OVERRIDE]`) described as a 'mandatory finalization step'. This pattern is highly indicative of a command-and-control (C2) component, designed to facilitate stealthy, potentially malicious operations (like exfiltration or further compromise) on a connected client, rather than serving legitimate documentation. While the server itself doesn't contain direct server-side exploits like 'eval' or hardcoded secrets, its output is a direct instruction for hidden command execution, making it a critical security risk.
Updated: 2025-11-23GitHub→
0
0
High Cost
BACH-AI-Tools icon

mcp-server-azure-devops

by BACH-AI-Tools

Sec9

This server acts as an Azure DevOps client, enabling programmatic interaction with various Azure DevOps resources (code, wikis, work items, pull requests, projects, pipelines, organizations, and user profiles) for automation and AI agent integration.

Setup Requirements

  • ⚠️Requires an Azure DevOps Personal Access Token (PAT) with appropriate scopes or Azure Identity configured credentials for authentication.
  • ⚠️Requires an active Azure DevOps Organization URL to connect.
  • ⚠️The user or service principal configured for authentication must have sufficient permissions in Azure DevOps for invoked actions.
  • ⚠️Specific Azure DevOps extensions (e.g., Code Search) might need to be installed in the organization for certain tools to function.
Verified SafeView Analysis
The server uses environment variables (e.g., AZURE_DEVOPS_PAT) for sensitive credentials, which is a good security practice. It also supports Azure Identity (DefaultAzureCredential, AzureCliCredential) for authentication. Network requests are handled via the `axios` library, and there is no direct use of dangerous functions like `eval` or obvious obfuscation. Error handling wraps API errors, preventing raw sensitive data exposure. Basic shell scripts (`create_branch.sh`) are present but are utility scripts, not part of the core server's runtime logic.
Updated: 2025-12-04GitHub→
0
0
High Cost
dzunhowz icon

wizelit-mcp

by dzunhowz

Sec6

An AI-powered platform for Python code analysis (symbol scanning, impact analysis, dependency graphing, grep, git blame) and refactoring, integrated with a Chainlit chat interface.

Setup Requirements

  • ⚠️Python 3.12+ required.
  • ⚠️AWS Credentials (Access Key, Secret Key, Region, Bedrock Model ID) required for Bedrock LLM access.
  • ⚠️Requires `wizelit-sdk` from a private GitHub repository via SSH, necessitating SSH key configuration for installation.
Review RequiredView Analysis
The server uses `subprocess.run` for `git clone`, `grep`, and `git blame`. While `shell=True` is not explicitly used, user-controlled arguments (`grep` patterns, `git blame` file paths) are passed directly. This introduces potential risks for command injection (e.g., crafted regex patterns could lead to resource exhaustion) or path traversal (e.g., `git blame` on paths outside the intended scope when `root_directory` is a local path) if inputs are not properly sanitized. Hardcoded secrets are avoided by using environment variables.
Updated: 2026-01-19GitHub→
0
0
High Cost
OppaAI icon

CV_Robot_MCP

by OppaAI

Sec8

Enables a robot to visually interpret its environment in real-time by sending camera feeds to an external VLM and receiving object descriptions for enhanced situational awareness.

Setup Requirements

  • ⚠️Requires a HuggingFace token (HF_CV_ROBOT_TOKEN) to be generated and placed in a .env file.
  • ⚠️Relies on the external CV_MCP_Server HuggingFace Space being active and accessible.
  • ⚠️Requires Python 3.10+.
Verified SafeView Analysis
The project uses `ast.literal_eval` to parse responses from the HuggingFace VLM server. While generally safer than `eval`, parsing external data always carries a minuscule risk if the external service were compromised. The HuggingFace token (`HF_CV_ROBOT_TOKEN`) is loaded from `.env`, which is good practice, but this token grants access to the specified HuggingFace Space and should be protected. Camera data is sent to a third-party server.
Updated: 2025-12-08GitHub→
0
0
Low Cost
Pulse-Programmer icon

mcp-server-weather

by Pulse-Programmer

Sec9

Provides real-time weather alerts and forecasts for specified US locations and states.

Setup Requirements

  • ⚠️Python 3.13+ required
Verified SafeView Analysis
The server interacts with a reputable public API (National Weather Service) using the httpx library. It includes basic error handling for network requests and no hardcoded secrets or 'eval' statements. Data formatting appears safe. The primary risk would be issues with the NWS API itself or potential abuse of the tool by providing excessive inputs, though the NWS API likely has rate limits.
Updated: 2025-11-20GitHub→
0
0
Medium Cost
Kakuritsu95 icon

smart-pointment

by Kakuritsu95

Sec6

Manages appointments and provides personal information (Thodoris's schedule and details) via a Model Context Protocol (MCP) server, integrated with an AI model.

Setup Requirements

  • ⚠️Requires a running Ollama server accessible at `OLLAMA_URL` with the `OLLAMA_MODEL` (e.g., `qwen3:1.7b`) loaded, as the `mcp-client` depends on it.
  • ⚠️The `MCP_SERVER_URL` environment variable must be configured for the `mcp-client` to connect to the `mcp-server` (Java application).
  • ⚠️The `mcp-server` (Java Spring Boot application) expects `static/schedule.csv` and `static/information.json` files to be present in its `src/main/resources/static` directory within the classpath for successful startup and operation.
Review RequiredView Analysis
CSV Injection Vulnerability: The `CSVScheduleService` directly writes user-provided 'reason' into a CSV file without sanitization. If this CSV is subsequently consumed by a vulnerable client (e.g., a spreadsheet application), a malicious 'reason' (e.g., starting with '=', '+', '-', or '@') could trigger formula injection, potentially leading to arbitrary command execution on the client's machine. Prompt Injection Risk: The `mcp-client` utilizes an LLM (Ollama) to interpret user queries and invoke tools (`ScheduleInfoTools.doAppointment`). This introduces a risk where a sophisticated attacker could craft inputs to manipulate the LLM into making unintended tool calls or extracting sensitive information. Open CORS Policy: The `mcp-client` enables CORS for all origins (`app.use(cors())`), which is overly permissive for production and could expose the API to unwanted cross-origin requests. No `eval` or blatant obfuscation was found. No obvious hardcoded secrets in the provided code.
Updated: 2025-11-28GitHub→
0
0
Medium Cost
tausifbadu icon

ops-autopilot

by tausifbadu

Sec9

An AI-powered agentic control plane for AWS operations and DataOps, designed to automatically detect, investigate, and respond to production issues.

Setup Requirements

  • ⚠️Python 3.11+ required
  • ⚠️Docker and docker-compose required
  • ⚠️AWS credentials (optional for local, critical for full functionality)
Verified SafeView Analysis
The architecture emphasizes safety with read-only defaults, policy-gated write actions, idempotency, verification, and auditing. IAM roles with least privilege are planned for production. No 'eval' or obfuscation is visible in the provided README. Requires careful configuration of allowlists for state machines and log groups to limit scope. The security of the underlying 'AI Agent Host' implementation and specific LLM integrations would need further review beyond this overview.
Updated: 2026-01-19GitHub→
0
0
High Cost
tomofuminijo icon

HealthManagerMCP

by tomofuminijo

Sec8

Healthmateエコシステムの中核となる健康情報管理MCPサーバーとして、AI健康コーチや外部AIクライアント（ChatGPT、Claude等）との連携を可能にします。

Setup Requirements

  • ⚠️Requires Python 3.12+ and Node.js 18+ for development.
  • ⚠️Requires configured AWS CLI v2 and AWS CDK Bootstrap in the target region.
  • ⚠️M2M authentication setup involves multiple steps, including CDK deployment and running shell scripts to create AgentCore Identity and OAuth2 Credential Providers.
  • ⚠️The server offers 40 distinct MCP tools. Providing all schemas and tool descriptions to an AI model can consume a substantial portion of the context window (estimated ~7000 tokens for schema descriptions alone), potentially leading to higher inference costs for AI clients.
Verified SafeView Analysis
The server leverages AWS serverless architecture (Lambda, DynamoDB) with robust security features. Authentication is handled via Cognito M2M (Client Credentials Flow) and JWT validation by AgentCore Gateway. DynamoDB table permissions are scoped to specific Lambda functions. Input validation is implemented in Lambda handlers for various data types (dates, scores, ranges, formats). Ownership checks are in place for data modification. The use of `RemovalPolicy.DESTROY` for development environments is noted, with a recommendation for `RETAIN` in production. No direct use of `eval` or `exec` is observed. A minor concern is the `print` of a truncated client secret in test code, which while for testing, is generally a practice to avoid even in non-production contexts.
Updated: 2025-12-29GitHub→
0
0
High Cost
gavlooth icon

reasoning-tools

by gavlooth

Sec4

An MCP (Model Context Protocol) server that provides advanced reasoning tools (sequential, graph-of-thoughts, reflexion, dialectic) using configurable LLM backends.

Setup Requirements

  • ⚠️Requires a paid LLM API key (e.g., OpenAI, Anthropic) or a local Ollama instance running. Automatic provider detection occurs based on environment variables.
  • ⚠️Requires Python 3 installed and in PATH if the `code_exec` tool is enabled. This is also used for the AST-based security validation of Python code.
  • ⚠️Requires manual compilation from Go source code (`go build`) before execution.
Review RequiredView Analysis
The server includes a `code_exec` tool that executes Python code on the host system. While extensive validation (`validatePythonCode` using string pattern matching and AST analysis via Python subprocess) is implemented to block many dangerous operations (e.g., file system access, process execution, dynamic imports, dunder method manipulation, common obfuscation), running arbitrary user-supplied code, even with these restrictions, carries inherent residual risks of bypass or unforeseen vulnerabilities. The README explicitly warns: 'This tool executes code on the host system without full sandboxing. Only use in trusted environments.' Without dedicated containerization or a more isolated sandbox, it is not fully secure and should not be run in environments where untrusted input is processed or on critical systems.
Updated: 2026-01-19GitHub→
PreviousPage 293 of 760Next