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)

31
3
High Cost
anutechofficial icon

simple-agent

by anutechofficial

Sec8

This project implements a Micro-Plugin (MCP) server that exposes tools for an AI agent, allowing it to collect structured user data via conversation.

Setup Requirements

  • ⚠️Requires `GITHUB_TOKEN` (or `OPENAI_API_KEY`) environment variable for the agent to function.
  • ⚠️Requires `ts-node` to be installed (via npm) to run TypeScript files directly.
  • ⚠️The MCP server (`server/index.ts`) and the AI agent (`agent/main.ts`) must be run separately (e.g., in different terminals) for the full system to operate.
Verified SafeView Analysis
The server uses `cors()` without specific origin restrictions, which is acceptable for local development but poses a security risk if exposed publicly. Tool inputs from the agent are used directly but within defined logical paths without apparent code injection vulnerabilities. Sensitive API keys like `GITHUB_TOKEN` are loaded securely via environment variables.
Updated: 2025-11-20GitHub→
31
1
High Cost
reasonkit icon

reasonkit-web

by reasonkit

Sec8

A high-performance Model Context Protocol (MCP) server for browser automation, web capture, and content extraction, enabling AI agents to interact with web browsers.

Setup Requirements

  • ⚠️Requires installation of Chrome/Chromium for headless browser automation.
  • ⚠️Requires `wasm-pack` for WebAssembly-related tests/builds (not strictly for server runtime).
  • ⚠️Relies heavily on environment variables for sensitive (tokens) and operational configuration; missing variables can lead to errors or insecure defaults.
  • ⚠️The primary HTTP server (defined in `src/main.rs`) is noted as a 'placeholder' and may not be fully functional, while the stdio-based MCP server (`src/bin/mcp_server.rs`) is complete.
Verified SafeView Analysis
The server demonstrates strong security awareness through its use of environment variables for all secrets, constant-time comparison for authentication tokens (though the token hashing itself uses a non-cryptographic hasher, which is a minor defense-in-depth point), strict localhost-only CORS by default, and a robust rate-limiting system. It also includes explicit warnings about binding to all interfaces and logging sensitive payloads. The primary security risk is the 'web_execute_js' tool, which allows arbitrary JavaScript execution within the browser. While this is an intended powerful feature for AI agents, it mandates that the MCP server must be exposed only to trusted clients/environments. Default configurations (localhost binding, authentication required for tools) mitigate this for typical deployments.
Updated: 2026-01-19GitHub→
31
1
High Cost
victor20252025 icon

opencode-config

by victor20252025

Sec2

Orchestrates a multi-agent system to decompose, execute, and review complex software development tasks in parallel using Git worktrees.

Setup Requirements

  • ⚠️Requires Bun runtime for execution.
  • ⚠️Requires Git CLI installed and available in PATH.
  • ⚠️The `repo-explorer` tool requires external CLI tools: `tree`, `ripgrep` (rg), `ast-grep` (sg), and `fd-find` (fd) to be installed.
  • ⚠️This is an OpenCode AI plugin, requiring the OpenCode AI platform or application to host and run it.
Review RequiredView Analysis
The `repo_search` and `repo_ast` tools in `tool/repo-explorer.ts` are highly vulnerable to command injection. User-provided `pattern` and `fileGlob` arguments are directly interpolated into shell commands executed via `sh -c` without apparent sanitization or escaping. An attacker could inject arbitrary shell commands by crafting malicious input for these parameters. For example, a `pattern` containing a single quote followed by `; rm -rf /;` would execute `rm -rf /`. This is a critical security flaw.
Updated: 2026-01-19GitHub→
31
1
Medium Cost
avinashsingh icon

mcp-sdlc-tracker

by avinashsingh

Sec8

Provides a Model Context Protocol (MCP) server for SQLite-based task and project tracking with full SDLC entity, wiki, and comments management.

Setup Requirements

  • ⚠️Requires Node.js installed locally to run `tsx`.
  • ⚠️Requires explicit initialization via the `initialize` MCP tool or `/api/initialize` endpoint, providing the current project directory path, before other tools can be used.
  • ⚠️Requires file system write permissions in the specified project directory to create the SQLite database file (`.project_tracker.donottouch`) and update `.gitignore`.
  • ⚠️The `get_knowledge_graph` tool depends on the `find` command, which is standard on Unix-like systems but might require specific setup on Windows.
Verified SafeView Analysis
SQL injection risks are largely mitigated by `better-sqlite3`'s consistent use of parameterized queries throughout the server. The `tools/write-full-file.ts` tool allows writing to arbitrary file paths within the initialized project directory, posing a moderate risk for an AI agent capable of arbitrary file write operations, though backups are created. The `tools/kg.ts` utilizes `child_process.execSync` but with hardcoded and well-defined `find` commands, limiting direct RCE from user input originating from client requests.
Updated: 2025-12-09GitHub→
31
1
Low Cost
ramblingenzyme icon

vscode-refactor-mcp

by ramblingenzyme

Sec8

A VSCode extension hosting an IPC server to enable external MCP (Model Context Protocol) servers to proxy VSCode API calls for refactoring and code manipulation.

Setup Requirements

  • ⚠️Requires Node.js v20+ and npm v9+ for development and building.
  • ⚠️Requires VSCode v1.90.0+ to run the extension and host the IPC server.
  • ⚠️The MCP server (`packages/server`) requires the `VSCODE_MCP_SOCKET_PATH` environment variable to connect, which is dynamically generated and set by the VSCode extension upon activation. Running the server independently requires manual setup of this variable.
  • ⚠️Uses npm workspaces; building the entire project is recommended via `npm run build` from the root.
Verified SafeView Analysis
The system utilizes local Inter-Process Communication (Unix Domain Sockets/Named Pipes) for communication between the VSCode extension and the MCP server, limiting external network exposure. Socket paths are dynamically generated per workspace, enhancing isolation. Data is exchanged as JSON strings, which are parsed by the extension. While standard, if a malicious MCP server were to connect and send malformed JSON, it could potentially lead to unexpected behavior or crashes. The primary security boundary relies on the MCP server being a trusted component of the system, as it instructs the extension to perform sensitive VSCode API calls like file renames and symbol renames. No 'eval' or obvious hardcoded secrets were found.
Updated: 2025-12-11GitHub→
31
1
Medium Cost
i3T4AN icon
Sec9

A personal semantic search engine and knowledge base application for various document types, offering natural language search, auto-clustering, 3D visualization, and AI agent integration.

Setup Requirements

  • ⚠️Requires Docker and Docker Compose for the recommended deployment method.
  • ⚠️Requires Python 3.11+ for native backend execution.
  • ⚠️Node.js 18+ is required to set up the MCP integration with AI agents like Claude Desktop.
  • ⚠️The embedding model (~400MB) is downloaded automatically on the first run, which may take several minutes.
Verified SafeView Analysis
The project demonstrates good security practices including filename sanitization to prevent path traversal during file uploads, configurable CORS, and rate limiting on critical endpoints. The destructive `/reset` endpoint is protected by an optional `ADMIN_KEY` (which defaults to empty, requiring user awareness to enable protection). The Model Context Protocol (MCP) integration explicitly excludes dangerous operations like file uploads and the `/reset` command from AI agent access, and supports OAuth for production use, which is a strong positive.
Updated: 2025-12-11GitHub→
31
1
Low Cost
DocHatty14 icon
Sec9

Augments LLM capabilities for non-linear, associative thinking to enhance creative problem-solving, brainstorming, and divergent exploration of concepts.

Setup Requirements

  • ⚠️Requires Node.js and npm/npx to run.
  • ⚠️Designed to be integrated with an MCP-compatible LLM client (e.g., Claude Desktop, VS Code extension) via standard I/O (stdin/stdout), rather than being a standalone web service.
Verified SafeView Analysis
The server primarily uses stdio for communication as configured, limiting direct network exposure. Its core logic involves string processing and state management, without apparent use of 'eval' or other direct code execution vulnerabilities from user input. Dependencies include standard web frameworks (express) via the MCP SDK, but this specific server is configured for stdio. No hardcoded secrets were found.
Updated: 2025-12-09GitHub→
31
1
High Cost

mcp-nvidia

by bharatr21

Sec9

Empowers Large Language Models (LLMs) to perform targeted searches across NVIDIA's official domains, providing relevant and structured information.

Setup Requirements

  • ⚠️Requires Python 3.10+.
Verified SafeView Analysis
The server makes external network requests via the `ddgs` package to perform searches. However, it implements strong security controls including domain whitelisting (only nvidia.com and nvidia.github.io are allowed), input validation, and rate limiting, mitigating common risks. There is no indication of arbitrary code execution beyond its intended search function.
Updated: 2025-11-17GitHub→
31
1
Medium Cost
xcollantes icon

jira-mcp

by xcollantes

Sec7

The Jira MCP server provides a Model Context Protocol (MCP) interface, enabling Large Language Models (LLMs) to interact with Jira for tasks like searching, creating, updating, and managing tickets and sprints.

Setup Requirements

  • ⚠️Requires `jira-cli` to be installed and initialized separately on the host machine.
  • ⚠️Requires `JIRA_API_TOKEN` and `JIRA_AUTH_TYPE` environment variables to be set for `jira-cli` authentication, with potential inheritance issues for GUI-based LLM clients if not explicitly passed.
  • ⚠️Python 3.12+ required (specified in `pyproject.toml`).
Verified SafeView Analysis
The server executes `jira-cli` commands via `subprocess.run`. While `subprocess.run` itself is not inherently unsafe, the direct passing of arguments, including JQL queries (via `--jql`), from LLM outputs to `jira-cli` introduces a potential for 'JQL injection' if the LLM is prompted to craft malicious JQL. This relies on `jira-cli`'s internal sanitization. Credentials (`JIRA_API_TOKEN`, `JIRA_AUTH_TYPE`) are handled via environment variables, which is a good practice. No `eval` or hardcoded secrets found in the Python source.
Updated: 2026-01-18GitHub→
31
1
Medium Cost
marlonluo2018 icon
Sec9

Integrate AI assistants with Microsoft 365 services, including email, calendar, contacts, files, and Teams, via the Microsoft Graph API.

Setup Requirements

  • ⚠️Requires interactive device code flow authentication, where the user must visit a URL and enter a code in a browser to complete login.
  • ⚠️Requires Python 3.10+.
  • ⚠️Dependencies must be installed using pip or uv.
Verified SafeView Analysis
Uses Microsoft's public client ID by default, configurable via .env for custom app registrations. Employs device code flow for authentication, which is user-interactive and avoids exposing client secrets. Authentication tokens are managed by MSAL and stored locally in the user's home directory, which is common for user-local applications. The project includes an input validation layer to prevent invalid inputs. Network requests are made to legitimate Microsoft Graph API endpoints, with rate limiting and exponential backoff implemented to handle API limits. No 'eval' or obvious obfuscation patterns were found.
Updated: 2026-01-19GitHub→
31
1
Low Cost
justmejulian icon

GardenaSmartSystem-MCP

by justmejulian

Sec8

Integrate with Gardena smart system API to monitor and control smart garden devices via an MCP server.

Setup Requirements

  • ⚠️Requires Java 21 runtime.
  • ⚠️Requires GARDENA API credentials (Client ID and Client Secret) obtained from the Husqvarna developer portal.
  • ⚠️Requires setting environment variables GARDENA_CLIENT_ID and GARDENA_CLIENT_SECRET.
  • ⚠️Requires building a standalone JAR (`./gradlew build`) before running with Claude Desktop, creating `app/build/libs/app-all.jar`.
Verified SafeView Analysis
Credentials (Client ID, Client Secret) are correctly loaded from environment variables. Ktor client is used for API communication. No 'eval' or direct external process execution on user input. The main security consideration is ensuring environment variables and the Claude Desktop configuration file are properly secured by the user.
Updated: 2025-11-24GitHub→
31
1
High Cost
kpeacocke icon

souschef

by kpeacocke

Sec2

AI-powered Model Context Protocol (MCP) server for comprehensive Chef-to-Ansible migration, including cookbook analysis, resource conversion, Habitat-to-Docker transformation, InSpec profile translation, and CI/CD pipeline generation.

Setup Requirements

  • ⚠️Requires Python 3.14+ (a very recent Python version, might require specific installation/management).
  • ⚠️Multiple critical path injection and archive extraction (Tar Slip) vulnerabilities are currently open, making it unsafe to process untrusted files or paths in a production environment.
  • ⚠️Requires an MCP-compatible client (e.g., Claude Desktop, VS Code 1.102+ with GitHub Copilot extension) for AI-assisted workflows. Commercial AI models will incur API costs.
Review RequiredView Analysis
The project has multiple critical security vulnerabilities identified as 'open' by CodeQL and SnykCode. These include 'Arbitrary File Write via Archive Extraction (Tar Slip)' in `ui/pages/cookbook_analysis.py` (line 308) and numerous instances of 'Uncontrolled data used in path expression (Path Injection)' across `souschef/ui/pages/cookbook_analysis.py`, `souschef/core/path_utils.py`, and `souschef/assessment.py`. These vulnerabilities could allow an attacker to write arbitrary files or access unexpected resources on the host system if user-provided input, particularly archive files, is processed. While the project demonstrates security awareness (CodeQL setup, explicit code standards), the presence of several critical unaddressed path manipulation issues significantly reduces its safety for processing untrusted inputs.
Updated: 2026-01-19GitHub→
PreviousPage 198 of 760Next