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
Medium Cost

mcp-optimization

by Foodcoman

Sec7

Provides documentation and strategic plans for optimizing development workflows, automation, and infrastructure across various tools and platforms, with a significant focus on AI/LLM integration.

Setup Requirements

  • ⚠️Requires access to an LLM API (e.g., Anthropic Claude API Key, likely a paid service) for AI-related tasks.
  • ⚠️A Docker environment is potentially required for implementing the described infrastructure plans.
  • ⚠️An n8n instance is likely required for executing the workflow automation plans.
  • ⚠️iTerm2 (and thus macOS) may be required if the 'iterm2-automation' directory contains specific scripts.
Verified SafeView Analysis
The repository includes a 'credentials-security-plan.md' indicating an awareness of security best practices. However, actual executable code is not available for review to assess direct risks like 'eval' or obfuscation. The plans for Chrome, n8n, and iTerm2 automation suggest potential for sensitive operations or external API interactions, which would require secure implementation if these plans are put into practice. The listed files themselves are primarily documentation and pose no direct execution risk.
Updated: 2025-11-17GitHub→
0
0
Low Cost
AuraFriday icon
Sec9

Provides a local, private, and automatically cached service for generating 1024-dimensional Qwen3-Embedding-0.6B vectors from text, supporting over 100 languages for semantic search and RAG systems.

Setup Requirements

  • ⚠️Requires an initial ~600MB model download from HuggingFace Hub on the first run (takes 2-5 minutes, requires internet).
  • ⚠️Automatic dependency installation via 'pip' requires 'pip' to be available and write permissions in the environment.
  • ⚠️Requires approximately 2GB of RAM for the Qwen model.
Verified SafeView Analysis
The server's core design emphasizes local inference and privacy, with no API calls sending user data externally after the initial model download. The `TOOL_UNLOCK_TOKEN` is a dynamically generated security measure for AI interaction, not a hardcoded secret. SQLite usage employs parameter binding, preventing SQL injection. The primary network risk is the one-time, automatic download of the Qwen3-Embedding-0.6B model (~600MB) and Python dependencies (`sentence-transformers`, `transformers`) from trusted sources (HuggingFace Hub, PyPI) during the first run. While `pip.main` is used for auto-installation, it targets known libraries. No 'eval' or obvious obfuscation is present.
Updated: 2025-12-02GitHub→
0
0
Low Cost
torkian icon

rag-mcp-server

by torkian

Sec9

Enables Claude Desktop to perform Retrieval-Augmented Generation (RAG) by connecting to a custom knowledge base via the Model Context Protocol (MCP).

Setup Requirements

  • ⚠️Requires Claude Desktop application.
  • ⚠️Python 3.10 or higher is required.
  • ⚠️Requires specifying absolute file paths for the server script and Python executable in Claude Desktop's configuration.
Verified SafeView Analysis
The provided source code for the MCP server is straightforward and does not contain obvious security vulnerabilities such as 'eval' or hardcoded secrets. It communicates via standard I/O (stdio_server). The primary security considerations would arise from how users customize the 'search_knowledge_base' function, e.g., if they integrate with insecure external APIs or handle sensitive data improperly. The base code itself is safe.
Updated: 2026-01-16GitHub→
0
0
Low Cost
thorsteneckel icon

nocodb-mcp-server-builds

by thorsteneckel

Sec10

This repository provides a CI/CD pipeline for building and publishing Docker images of the NocoDB MCP Server.

Setup Requirements

  • ⚠️Requires Docker to be installed and running on the host system.
  • ⚠️Requires an existing NocoDB instance with a valid API token and Base ID for the server to connect and function properly.
  • ⚠️Mandatory environment variables (NOCODB_URL, NOCODB_API_TOKEN, NOCODB_BASE_ID) must be provided during container runtime for proper configuration.
Verified SafeView Analysis
The provided "SOURCE CODE" is solely the `README.md` of the `nocodb-mcp-server-builds` repository. This README describes a CI/CD pipeline for building Docker images and how to run them. It does not contain executable code, server logic, `eval` statements, obfuscation, or direct network calls within itself. It correctly instructs users to pass sensitive information (like `NOCODB_API_TOKEN`) via environment variables. The security score reflects only the content of the provided `README.md` and not the underlying `edwinbernadus/nocodb-mcp-server` application or the GitHub Actions workflow files which were not provided.
Updated: 2025-12-02GitHub→
0
0
Medium Cost
gnufoo icon

MeCP

by gnufoo

Sec8

A self-hosted Rust server for AI context management, offering modular integrations with databases and LLM reasoning via the Model Context Protocol.

Setup Requirements

  • ⚠️Installation scripts for MySQL and Neo4j are currently designed for WSL Ubuntu/Debian systems.
  • ⚠️Milvus integration (if enabled) requires Docker for local deployment.
  • ⚠️Leveraging external LLM providers (e.g., OpenAI, Anthropic) for reasoning will incur API costs and requires API keys.
Verified SafeView Analysis
The server implements robust Web3 authentication with JWT sessions, daily nonces, and signature verification, along with address whitelisting. Key configuration values like database passwords and JWT secrets are defaulted in `config.toml` but clearly documented for change in production, with helper scripts to generate secure secrets. `CorsLayer::permissive()` is used, which is permissive for development but a risk in production without explicit configuration. The Neo4j installation script sets `server.default_listen_address=0.0.0.0` which can expose the database if not behind a firewall. However, the documentation explicitly highlights hardening for production. No obvious malicious patterns or 'eval' usage.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
Dub1n icon

stelae

by Dub1n

Sec6

Transforms a local WSL workspace into a single, extensible MCP endpoint for desktop agents, ChatGPT Connectors, and other HTTP/SSE clients, enabling declarative tool overrides and aggregations for AI agent tooling.

Setup Requirements

  • ⚠️Requires WSL with systemd enabled.
  • ⚠️Requires Python 3.11+, Go toolchain, Node.js (via NVM recommended), `pm2`, and `cloudflared` CLI.
  • ⚠️Extensive configuration via environment variables and `.env` files (e.g., STELAE_CONFIG_HOME, PUBLIC_BASE_URL) is crucial for proper operation.
  • ⚠️Path sensitivity: workspaces must be on WSL's ext4 filesystem, not Windows-backed mounts (e.g., /mnt/c/).
Review RequiredView Analysis
The server is designed to execute local commands, manage filesystems, and install/remove other MCP servers, which exposes a significant attack surface if not properly secured at the deployment and access control layers. It relies heavily on `subprocess.run` and `urllib.request.urlopen` for core functionalities and external interactions. Configuration changes via `manage_stelae` allow the server to self-modify its operational capabilities. While the code includes some path validation and environment variable expansion, robust external authentication and authorization are critical, especially when exposed publicly via Cloudflare. Hardcoded secrets are generally avoided, with `OPENAI_API_KEY` and `GITHUB_TOKEN` being passed via environment variables.
Updated: 2025-11-25GitHub→
0
0
Low Cost
Rbfinch icon

hindsight-mcp

by Rbfinch

Sec9

Provides AI assistants with searchable development history (git commits, test results, GitHub Copilot sessions) via Model Context Protocol (MCP) tool calls in VS Code.

Setup Requirements

  • ⚠️Requires VS Code v1.99+ with GitHub Copilot.
  • ⚠️For test ingestion, `cargo-nextest` must be installed separately via `cargo install cargo-nextest`.
Verified SafeView Analysis
Robust against common injection attacks due to parameterized SQL queries using `rusqlite::params![]` and extensive fuzz testing on input parsing. Command execution (`std::process::Command`) is used for `cargo nextest`, which generally handles arguments safely, but direct execution of external processes always carries a minimal inherent risk. The VS Code extension downloads binaries from GitHub Releases, relying on GitHub's integrity for secure distribution.
Updated: 2026-01-19GitHub→
0
0
High Cost
you0best icon

my-mcp-server

by you0best

Sec8

This project serves as a TypeScript boilerplate for developing Model Context Protocol (MCP) servers, providing various AI-enabled and utility tools for integration with AI agents.

Setup Requirements

  • ⚠️Requires a Hugging Face API Token (`HF_TOKEN`) for the image generation functionality.
  • ⚠️Uses the Smithery CLI for building and development (`npm run build`, `npm run dev`).
  • ⚠️Requires Node.js version 18 or higher.
Verified SafeView Analysis
The server includes an image generation tool that relies on an external Hugging Face API. The API token (`hfToken`) is loaded from configuration, which is a good practice, preventing hardcoded credentials. No direct 'eval' or malicious code execution patterns were found. The 'code_review' prompt generates text for an external LLM, not executing code locally. The `express-rate-limit` dependency in `package-lock.json` suggests network security considerations, but the core server logic provided here uses a `StdioServerTransport` in the example, which is generally safer for local process communication. Overall, the server appears robust for a boilerplate, but users must ensure their `hfToken` is secured and be aware of potential large data transfers for image generation.
Updated: 2025-11-27GitHub→
0
0
High Cost
antonsokolskyy icon

SSMCP

by antonsokolskyy

Sec7

Model Context Protocol (MCP) server providing web search with content extraction and YouTube subtitle retrieval for AI models lacking built-in browsing capabilities.

Setup Requirements

  • ⚠️Requires Docker and Docker Compose for deployment.
  • ⚠️Requires manual configuration of SearXNG (editing `deploy/docker/searxng_data/settings.yml` to add `json` format) after initial setup, before building the main image.
  • ⚠️Optional: Providing YouTube cookies (in Netscape format) for restricted content requires placing a `cookies.txt` file in a specific directory (`deploy/docker/ssmcp/cookies.txt`) and setting appropriate file permissions (`chmod 600`).
  • ⚠️If OAuth is enabled (`OAUTH_ENABLED=true`), specific environment variables (`OAUTH_JWKS_URL`, `OAUTH_CLIENT_ID`, `OAUTH_ISSUER`) must be configured correctly for an OIDC-compliant identity provider, including the OpenWebUI client ID for audience verification.
Verified SafeView Analysis
The server uses `httpx` for external requests and `pydantic-settings` for configuration, reducing risks of hardcoded secrets. OAuth authentication is implemented using standard JWT validation (JWKS, issuer, audience, expiration, subject), which is generally secure when configured with trusted identity providers. However, the reliance on external tools like `Crawl4AI` (headless Chromium) and `yt-dlp` for arbitrary web content introduces a significant attack surface. Specifically, `Crawl4AI`'s `ignore_https_errors=True` setting weakens HTTPS certificate validation, making connections vulnerable to Man-in-the-Middle (MITM) attacks. While explicit warnings are provided for `cookies.txt` permissions, the inherent risk of processing untrusted web content via browser automation remains.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
leftmike icon

filemcp

by leftmike

Sec10

Provides an MCP interface for an AI agent to interact with a local filesystem, offering operations like reading files, listing directories, searching, and retrieving file metadata.

Setup Requirements

  • ⚠️Requires a Go runtime environment to build and execute from source.
  • ⚠️Communication is exclusively via standard I/O (stdin/stdout), requiring a parent process capable of interacting through this interface.
  • ⚠️The server operates strictly within a specified root directory, limiting its filesystem access.
Verified SafeView Analysis
The server uses `os.DirFS` to create a sandboxed filesystem, restricting all file operations to a specified root directory. This effectively prevents path traversal attacks (e.g., `../`, absolute paths) as confirmed by the provided test cases. There are no dynamic code execution features (like 'eval'), no apparent network attack surface (it uses StdioTransport for communication), and no hardcoded sensitive information.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
cosmocoder icon

mcp-web-docs

by cosmocoder

Sec9

A self-hosted MCP server that crawls, indexes, and enables private, local search of documentation from any website to provide context to AI assistants.

Setup Requirements

  • ⚠️Requires Node.js >= 22.19.0.
  • ⚠️Initial setup requires downloading Playwright browsers (Chromium, Firefox, WebKit) which consumes disk space and time.
  • ⚠️Indexing large documentation sites consumes significant local CPU, RAM, and disk space for vector database and crawl cache.
  • ⚠️Requires manual configuration in your MCP client (e.g., Cursor, Claude Desktop, VS Code) to connect to the server.
  • ⚠️For private documentation, an interactive browser login is required to capture the session.
Verified SafeView Analysis
Comprehensive security measures implemented: SSRF protection (blocks private IPs, localhost, cloud metadata), ReDoS protection for user-provided regex, robust Zod schema validation for all inputs, AES-256-GCM encryption for stored authentication sessions, sensitive data redaction in logs/errors, and proactive prompt injection detection in crawled content (high-severity patterns blocked, others warned). No discernible hardcoded secrets or arbitrary `eval` usage.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
abhinav20sep icon

gmail-mcp-server-rust

by abhinav20sep

Sec4

Integrate AI models (e.g., Claude) with Gmail for email management and automation via the Model Context Protocol.

Setup Requirements

  • ⚠️Requires Rust 1.70+ installed via rustup.
  • ⚠️Requires a Google Cloud Project with Gmail API enabled and OAuth 2.0 Desktop app credentials.
  • ⚠️Requires an interactive, browser-based authentication step ('./gmail-mcp-server auth') before the server can be used.
Review RequiredView Analysis
The `download_attachment` tool allows the AI (and thus potentially an attacker through a crafted prompt) to specify an arbitrary `save_path` and `filename`. While the `Path::join` method correctly handles path segments to prevent simple path traversal within the filename, it does not validate or restrict the base `save_path` argument. This means an attachment could be downloaded to sensitive system directories (e.g., `/etc`, `~/.ssh`) if the AI is prompted to do so, posing a significant risk for arbitrary file writes.
Updated: 2025-11-26GitHub→
PreviousPage 320 of 760Next