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
1
Low Cost
mugdhav icon

mcpservers

by mugdhav

Sec9

Provides a privacy-first, local AI-powered semantic search engine for local media (images and videos) to AI assistants like Claude Desktop, enabling natural language queries for media organization and retrieval.

Setup Requirements

  • ⚠️Requires significant initial download for PyTorch (~2GB) and the SigLIP AI model (~150MB).
  • ⚠️Requires manual editing of Claude Desktop's `claude_desktop_config.json` with absolute paths, which can be error-prone, especially on Windows.
  • ⚠️The server needs user-provided media files in the configured `MEDIA_DIR` to be functional; it does not come with sample media.
Verified SafeView Analysis
The server's core functionality involves processing local media files and does not send any data to external cloud services. It uses standard libraries (transformers, PyTorch, OpenCV, FAISS). While the `get_media_details` function accepts a `file_path` argument, within the context of an MCP server designed for local interaction and indexed content, this is generally considered safe. There are no apparent uses of `eval`, unsanitized shell commands, hardcoded credentials, or obfuscation. Model downloads from Hugging Face are standard and generally trusted.
Updated: 2025-12-08GitHub→
31
1
Low Cost

MCP-Http-Client

by ferPrieto

Sec7

This repository provides an HTTP client library or application designed to interact with services likely related to the Minecraft Mod Coder Pack (MCP) ecosystem.

Setup Requirements

  • ⚠️Requires Java Development Kit (JDK) installed.
  • ⚠️Configuration via 'example-config.json' or similar methods will likely be required for proper operation.
Verified SafeView Analysis
No source code was provided for detailed analysis of 'eval' usage, obfuscation, or specific network vulnerabilities. As a client, risks primarily involve the trustworthiness of the servers it connects to and how it handles received data.
Updated: 2025-11-18GitHub→
31
45
Low Cost
promplate icon

hmr

by promplate

Sec8

Provides a fine-grained hot module reloading (HMR) engine for Python applications like FastAPI, Uvicorn, and Model Context Protocol (MCP) servers, including a browser auto-refresh feature for development.

Setup Requirements

  • ⚠️Requires Python 3.12 or newer.
  • ⚠️When using `fastapi-reloader` manually, `html_injection_middleware` must be placed before any compression middleware (e.g., `GZipMiddleware`).
  • ⚠️Manual triggering of reload signals (e.g., monkey-patching `uvicorn.Server.shutdown`) is required for `fastapi-reloader` standalone usage with non-`uvicorn-hmr` servers to ensure graceful shutdown.
  • ⚠️The target module for `mcp-hmr` (and `hmr` generally) must not be imported before calling `mcp_server(target)` or `hmr path/to/your/entry-file.py` to ensure hot-reloading functionality.
Verified SafeView Analysis
The project is a development tool that employs runtime code execution (via `compile` and `exec`) to achieve hot module reloading. This is inherent to its functionality and is applied to local source code files detected as changed, not arbitrary external input. Network risks include exposed ports for ASGI/MCP servers and a long-polling connection for browser refresh, which are standard for the respective server types. No hardcoded secrets or malicious patterns were identified. The tool is safe for its intended development-only use.
Updated: 2025-12-30GitHub→
31
1
Medium Cost
williamzujkowski icon

nexus-agents

by williamzujkowski

Sec9

Coordinates multiple AI experts with model diversity and workflow automation to handle software development tasks.

Setup Requirements

  • ⚠️Requires Node.js v22.x LTS and pnpm v9.x (or npm v10.x).
  • ⚠️At least one AI provider API key (e.g., ANTHROPIC_API_KEY) is required for full AI functionality.
  • ⚠️GitHub CLI (gh) is recommended for PR review workflows.
Verified SafeView Analysis
The server adopts a 'security-first design' with multiple layers of defense, including Zod schemas for input validation, a secrets vault to prevent API key exposure, rate limiting, and robust path safety with directory jails. It explicitly prevents user-provided regular expressions to mitigate ReDoS attacks and implements execution sandboxing using policy-based or Docker container isolation. Active Byzantine detection is also in place for multi-agent decisions. Code reviews actively check for security vulnerabilities, demonstrating a proactive security posture.
Updated: 2026-01-19GitHub→
31
1
Low Cost
dock-ai icon

mcp

by dock-ai

Sec9

Allows AI agents to discover and interact with real-world businesses by resolving their domains to MCP connectors for actions like booking or shopping.

Setup Requirements

  • ⚠️Requires `INTERNAL_API_KEY` to be set, especially in production, to authenticate with the Dock AI backend.
  • ⚠️Relies on `DOCKAI_API_URL` for all backend API interactions (defaults to `https://api.dockai.co`).
  • ⚠️Rate limiting functionality requires an Upstash Redis instance and corresponding `UPSTASH_REDIS_REST_URL` and `UPSTASH_REDIS_REST_TOKEN` environment variables.
Verified SafeView Analysis
The server implements rate limiting with IP spoofing protection, delegating to Upstash Redis, and fails closed in production if not configured. OAuth 2.1 is implemented, but all sensitive client/user data and token management are delegated to an external 'dockai-api', reducing the local attack surface. Critical secrets like `INTERNAL_API_KEY` are retrieved from environment variables and are explicitly checked as required in production. No 'eval' or similar dangerous patterns are used directly within the provided source.
Updated: 2026-01-19GitHub→
31
1
Medium Cost
kiraboibrahim icon

askari-patrol-mcp-server

by kiraboibrahim

Sec8

AI-powered security management assistant for WhatsApp and CLI, interacting with the Askari Patrol / GuardTour API.

Setup Requirements

  • ⚠️Requires `GROQ_API_KEY` (paid API for LLM inference).
  • ⚠️Requires Twilio account credentials (`TWILIO_ACCOUNT_SID`, `TWILIO_AUTH_TOKEN`, `TWILIO_WHATSAPP_NUMBER`) for WhatsApp integration.
  • ⚠️Needs `ngrok` or similar for exposing the WhatsApp client webhook to the internet during local development.
  • ⚠️Requires Python 3.12 or newer.
Verified SafeView Analysis
All sensitive credentials (API keys, tokens) are correctly loaded from environment variables. No 'eval' or malicious patterns were found in the provided source code. External network calls are to known services (Askari Patrol API, Twilio). The JWT validation function properly handles signature verification and expiration.
Updated: 2025-11-29GitHub→
31
1
Low Cost
Ishaan300104 icon

multi-agent-reviewer

by Ishaan300104

Sec8

A multi-agent system designed to review, analyze, and summarize academic research papers by leveraging LLM-based agents and PDF/ArXiv processing.

Setup Requirements

  • ⚠️Requires OpenAI API Key (Paid service for LLM integration)
  • ⚠️Requires Python 3.9+
  • ⚠️Internet access is mandatory for ArXiv API and PDF downloads
  • ⚠️Multiple MCP server processes (4 in total) must be started simultaneously, although a convenience script is provided.
Verified SafeView Analysis
The system avoids direct use of 'eval' or similar dangerous functions on user input. API keys are managed via environment variables. The MCP servers bind to '0.0.0.0' and use permissive CORS ('allow_origins=["*"]'), which could pose a risk if deployed in a public-facing, non-isolated environment. However, for its intended local, internal multi-agent communication, this is a common development setup.
Updated: 2025-11-25GitHub→
31
1
Low Cost
Gitmaxd icon
Sec9

Facilitates interaction between MCP-compatible AI clients and Context Repo for managing prompts, documents, and collections, including semantic search.

Setup Requirements

  • ⚠️Requires Node.js 18 or higher.
  • ⚠️Requires an MCP-compatible client application (e.g., Claude Desktop, Cursor IDE).
  • ⚠️Requires a Context Repo account and a valid API key with appropriate permissions.
Verified SafeView Analysis
The server uses environment variables for API key authentication, which is a secure practice. It relies on the `@modelcontextprotocol/sdk` for handling MCP communication and validates inputs via JSON schemas (implied by Zod dependency). API requests to Context Repo are made using standard `fetch` with appropriate error handling for various HTTP statuses. User input is not directly executed or used in a way that suggests injection vulnerabilities. The build script uses `execSync`, but this is a build-time operation and not part of the runtime server logic handling user requests. The primary security considerations would be the robustness of the upstream Context Repo API and the secure management of the API key by the user.
Updated: 2025-12-15GitHub→
31
1
Medium Cost
bgtdevhub icon

fgdb-mcp-server

by bgtdevhub

Sec9

The FGDB MCP Server facilitates interaction with Esri File Geodatabases (FGDB) via the Model Context Protocol, enabling geospatial data querying, modification, and schema management through ArcPy.

Setup Requirements

  • ⚠️Requires ArcGIS Pro installation with ArcPy support (Windows OS only).
  • ⚠️Requires activation of the ArcGIS Pro conda environment to run (e.g., 'proenv.bat', 'conda activate arcgispro-py3').
  • ⚠️Installation requires running 'pip install -e .' possibly with Administrator privileges in the ArcGIS Pro conda environment.
Verified SafeView Analysis
The server demonstrates a strong commitment to security through a layered architecture, comprehensive input validation (e.g., `validate_where_clause` specifically checks for dangerous SQL patterns like 'DROP TABLE', '--', '/*'), and a mandatory safety confirmation system for all mutating operations (insert, update, delete, add/delete field). This confirmation mechanism requires a unique token, preventing direct, unconfirmed execution of high-risk commands. Path validation is present for GDB paths. The ArcPy backend calls are parameterized or use validated input, minimizing SQL injection risks. No direct use of 'eval' or 'exec' on user input was identified. Hardcoded secrets are not apparent. The `SECURITY.md` file outlines a clear policy and best practices. A perfect 10 is difficult due to reliance on external ArcPy libraries, but the Python layer is well-fortified.
Updated: 2025-11-27GitHub→
31
1
Low Cost
abordage icon

awesome-mcp

by abordage

Sec10

A dynamically updated, curated, and ranked directory of Model Context Protocol (MCP) servers, clients, and frameworks for AI assistants.

Verified SafeView Analysis
The provided source code is for an awesome list, not an executable MCP server. It consists of Markdown and YAML files that describe other projects. It contains no directly executable code, 'eval' statements, obfuscation, or hardcoded secrets. The security risk for a user primarily relates to external projects linked, not the content of this repository itself. Therefore, running (i.e., merely accessing or viewing) the content of this repository is safe.
Updated: 2026-01-19GitHub→
31
1
Medium Cost
jackdark425 icon

aigroup-stata-mcp

by jackdark425

Sec6

A Stata integration server based on Model Context Protocol (MCP) that allows AI assistants to execute Stata scripts, fetch data information, and run statistical analysis.

Setup Requirements

  • ⚠️Requires Python 3.11 or higher.
  • ⚠️Requires Stata statistical software to be installed and accessible on the host machine; manual configuration of the `STATA_CLI` environment variable might be necessary if auto-detection fails.
  • ⚠️Requires a compatible Model Context Protocol (MCP) client (e.g., Claude Desktop, Cherry Studio) to interact with the server.
Verified SafeView Analysis
The server utilizes `subprocess.Popen` and `pexpect` with `shell=True` for executing Stata commands. While this is necessary for its core function, it introduces a potential risk of shell injection if LLM-generated Stata code or file paths contain malicious shell commands. File operations such as `read_file`, `write_dofile`, `append_dofile`, and `load_figure` allow the LLM to interact with arbitrary file paths. Although `mk_dir` employs `pathvalidate` for sanitization, other file-related tools do not show explicit path sanitization against traversal attacks, which could potentially allow access or modification of files outside the intended working directory if an LLM is compromised. No hardcoded secrets were identified.
Updated: 2025-12-04GitHub→
31
1
Low Cost
biocontext-ai icon

nucleotide_archive_mcp

by biocontext-ai

Sec8

Search and access RNA sequencing datasets from the European Nucleotide Archive (ENA), retrieve detailed metadata, find related publications, and generate download scripts for data files.

Setup Requirements

  • ⚠️Requires Python 3.11 or newer.
  • ⚠️Requires internet access to ENA and Europe PMC APIs.
Verified SafeView Analysis
The server primarily acts as an API wrapper for ENA and Europe PMC, using `httpx` for network requests. Query parameters for external APIs are constructed with careful string formatting and explicit quoting for values, reducing typical injection risks within the ENA query language. The `generate_download_script` tool creates bash scripts with `wget`/`curl` and `md5sum`. While the commands themselves are generally safe, the filenames and URLs are derived from ENA API responses. A theoretical risk exists if a compromised ENA API were to return maliciously crafted URLs or filenames that could lead to shell injection in the generated script, but this is a low probability given the nature of ENA. No `eval` or `exec` functions are used, and sensitive information like API keys are expected to be managed via environment variables.
Updated: 2025-11-28GitHub→
PreviousPage 202 of 760Next