Stop Searching. Start Trusting.

The curated directory of MCP servers, vetted for security, efficiency, and quality.

Browse Directory Why 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)

Low Cost

MCP-Server-Validator

by Starborn

Sec5

Validates MCP Servers against a predefined specification of requirements to ensure compliance.

Review RequiredView Analysis

Insufficient source code provided for a comprehensive security audit. Only the README.md file was available for review. No 'eval', obfuscation, network risks, hardcoded secrets, or malicious patterns could be identified or ruled out due to the lack of code.

Updated: 2026-01-17GitHub→

High Cost

med-mcp-server

by abumuslim1

Sec6

Analyzing medical PDF conclusions for inconsistencies and errors using a Yandex Cloud-powered AI assistant.

Setup Requirements

⚠️Requires Yandex Cloud API Key and Folder ID (Paid service for AI Studio)
⚠️PDF files must be placed in the configured PDF_ROOT directory (default: data/pdfs) for analysis
⚠️Requires Python 3.x and specific dependencies (dotenv, pypdf, openai, fastmcp, uvicorn)

Review RequiredView Analysis

Secrets (Yandex Cloud API Key/Folder ID) are loaded from environment variables, which is good practice. No hardcoded secrets or `eval` usage. However, the `pdf_id` argument in `analyze_pdf` and `read_pdf_text` is used directly in `PDF_ROOT / pdf_id` to construct a file path. If `pdf_id` is not properly sanitized (e.g., to prevent `../` or absolute paths) by FastMCP or the caller, this could lead to a directory traversal vulnerability, allowing access to arbitrary files outside the intended `PDF_ROOT` directory.

Updated: 2025-11-23GitHub→

Medium Cost

claude-code-setup

by Danservfinn

Sec4

Enables Claude Code agents to make HTTP requests to retrieve information from arbitrary web URLs, facilitating data collection and real-time information access for development tasks.

Setup Requirements

⚠️Requires Node.js runtime environment.
⚠️Requires active network connectivity.
⚠️Potential for resource exhaustion with very large web page fetches.
⚠️May require proxy configuration in restrictive network environments.

Verified SafeView Analysis

A web fetching capability inherently carries risks, including Server-Side Request Forgery (SSRF) if not rigorously configured with egress filtering, potential for Denial of Service (DoS) through large file downloads or excessive requests, and information leakage if sensitive data is fetched from internal or restricted networks. Without reviewing the actual server implementation code, assumptions on input sanitization, timeout enforcement, and sandboxing cannot be definitively made. Users should be aware of the potential for unintended network access or data exposure.

Updated: 2026-01-17GitHub→

Medium Cost

superops-mcp-server

by Sborgi

Sec8

Connects AI assistants like Claude to SuperOps.ai for managing IT operations, including tickets, clients, and assets.

Setup Requirements

⚠️Requires Node.js 18 or higher.
⚠️Requires an active SuperOps.ai account with API access and a generated API token.
⚠️Requires Claude Desktop or another MCP-compatible client to utilize the server's tools.
⚠️Requires manual configuration of SuperOps.ai API Token, Subdomain, and Datacenter as environment variables.

Verified SafeView Analysis

The server handles sensitive SuperOps.ai API tokens via environment variables, which is a standard and generally secure practice. No 'eval' or explicit obfuscation mechanisms are mentioned. Interacting with external APIs always carries inherent risks related to credential exposure or upstream service vulnerabilities.

Updated: 2025-11-17GitHub→

Medium Cost

gossisMCP

by MCPRUNNER

Sec3

Analyzes SSIS (SQL Server Integration Services) DTSX files, providing detailed insights into package structure, data flows, control flows, logging, performance, security, and best practices, with extensible plugin support and workflow automation.

Setup Requirements

⚠️Requires Go 1.19 or later.
⚠️Plugin compilation (`go build -buildmode=plugin`) generates OS-specific shared libraries (`.so` on Linux/macOS, `.dll` on Windows), and the `create_analysis_rule` feature, if fully implemented, would enable on-the-fly compilation of arbitrary Go code into plugins, posing a significant security risk.
⚠️Primarily tested on Windows 11; path handling and file system operations may require further validation on non-Windows platforms, potentially impacting full functionality, especially for specific tools like MSMQ analysis.

Review RequiredView Analysis

The `create_analysis_rule` tool's design permits passing Go code as a string, which is then intended to be compiled and loaded as a dynamic plugin. If the `buildPlugin` function (currently a no-op in the provided code) were fully implemented, this would introduce a critical Remote Code Execution (RCE) vulnerability. Additionally, while the README mentions plugin signature verification and sandboxing, the `LoadPlugin` implementation does not visibly enforce these security measures before dynamically loading shared libraries (`.so`/.dll files). This makes the system vulnerable to malicious plugins executing arbitrary code. Path traversal attacks are explicitly tested and mitigated in file path resolution utilities, and the server includes tools to detect hardcoded credentials and sensitive data patterns within SSIS packages, rather than storing them itself.

Updated: 2025-12-08GitHub→

Low Cost

mcp_server_test

by jvermeir

Sec6

A simple REST API service for storing and retrieving text documents, primarily used as a local test server and integrated with Model Context Protocol (MCP) plugins for AI interaction.

Setup Requirements

⚠️Requires uv package manager for dependency management and execution.
⚠️Requires Python 3.11 or newer.
⚠️Manual configuration of Claude desktop (`claude_desktop_config.json`) with absolute paths to `uv` and the plugin folder is required for MCP integration.
⚠️The server must be running on port 3000 for the MCP plugin to connect by default.

Verified SafeView Analysis

The Flask server (`test_server/server.py`) runs with `debug=True` by default, which is a significant security risk for anything beyond local development, as it can expose sensitive information and allow arbitrary code execution via the Werkzeug debugger. The MCP plugin (`plugins/rest_call_plugin/rest_call.py`) makes HTTP requests using `httpx`. While it limits the host to `http://localhost:3000`, an AI instructing it could still send arbitrary data as JSON bodies or headers to the local server, which is generally acceptable given its role as a plugin for an AI agent interacting with a local service.

Updated: 2025-12-13GitHub→

Low Cost

pycontextify

by tbrandenburg

Sec9

Provides a semantic search server with lightweight knowledge graph capabilities for diverse knowledge sources like codebases and documents, integrated via the MCP protocol for AI assistants.

Setup Requirements

⚠️Requires 'uv' package manager for installation and execution.
⚠️Requires Python 3.10 or newer (full test suite targets Python 3.13+).
⚠️Requires FAISS, sentence-transformers, and PDF processing libraries (PyMuPDF, langchain-community) which may need to be installed.
⚠️Embedding models (sentence-transformers) are downloaded on first run, requiring internet access and potentially causing a longer initial startup time.
⚠️Requires 8 GB RAM minimum (16 GB recommended for larger corpora) as embeddings and FAISS indexes reside in-process.

Verified SafeView Analysis

The code generally follows good security practices, using `pathlib` for path manipulation and validating inputs. Hardcoded secrets are explicitly avoided, with `openai_api_key` being read from environment variables. The `BootstrapService` for downloading index archives is a potential supply chain risk if the archive URLs are untrusted, as it downloads and extracts remote content. However, it includes checksum verification to ensure integrity, which is a mitigating factor. `subprocess.run` calls are controlled and do not appear vulnerable to injection from user input within the server itself. No `eval` or obvious obfuscation patterns were found.

Updated: 2025-11-20GitHub→

High Cost

tyomarkkinatori_mcp_server

by markusnieminen1

Sec8

An MCP server to query Finnish labor market job listings from the Työmarkkinatori API based on user-provided location and occupation criteria.

Setup Requirements

⚠️Requires Python 3.12+
⚠️Requires Claude Desktop installed and configured for MCP servers
⚠️Requires Docker installed for containerized deployment, or `uv` for Python dependency management
⚠️The tool `get_all_high_level_occupation_codes` actually returns 'low-level' occupation codes (from `ammatit_taso_3.json`), which is verbose and might consume more tokens than expected for a 'high-level' request.

Verified SafeView Analysis

The server includes good security practices such as input validation for codes and municipality names, using Pydantic models for API request structure with length constraints, and a timeout for external HTTP requests. When run via Docker, it uses secure container options like `--read-only` and `--cap-drop=ALL`. No direct 'eval' or hardcoded secrets were found. Data files are local and validated at startup. However, the external API call response content is not explicitly sanitized before being returned to the LLM, which could, in theory, contain malicious content if the Työmarkkinatori API were compromised (though highly unlikely for job listings).

Updated: 2025-11-25GitHub→

Low Cost

MCP_expense_tracer_remort

by sreehari-sreesunil

Sec10

Provides basic mathematical operations through an MCP server.

Setup Requirements

⚠️Requires Python 3.13 or newer

Verified SafeView Analysis

The code implements basic arithmetic and random number generation with no apparent vulnerabilities, 'eval' calls, hardcoded secrets, or complex interactions that could be exploited.

Updated: 2025-12-14GitHub→

Medium Cost

AIDevToolsHack

by RuslanLat

Sec7

An AI agent for managing emails and calendar events by integrating with Large Language Models (LLMs) via the Model Context Protocol (MCP).

Setup Requirements

⚠️Requires API keys and credentials for LLM, MCP, email (IMAP/SMTP), and CalDAV services, many of which are paid or require app-specific passwords.
⚠️Requires active external email and CalDAV accounts for full functionality.
⚠️Python 3.12+ is required for `mcpserver` and `ui` services, while `agent` specifies 3.10+ (recommend ensuring 3.12+ for all).
⚠️Docker and Docker Compose are necessary for the easiest setup and running of all components locally.

Verified SafeView Analysis

The server uses environment variables for sensitive credentials (email, CalDAV, API keys), which is a good practice. It relies on standard, well-maintained libraries for network interactions (IMAP, SMTP, CalDAV). The email body cleaning function `_clean_email_body` is complex with multiple regex operations, which can be prone to edge cases, but aims to reduce the amount of data sent to the LLM. The primary security consideration lies in giving an LLM access to email and calendar actions; misconfigured or unconstrained prompts could lead to unintended email sending or event creation. No direct `eval` or blatant malicious patterns were found. However, interaction with external LLMs and their data handling policies should be considered for privacy.

Updated: 2025-12-12GitHub→

Low Cost

mcpServer

by gharshit

Sec3

Provides a standardized interface for AI assistants to interact with PostgreSQL databases securely and contextually.

Setup Requirements

⚠️Requires an existing PostgreSQL database instance.
⚠️Database connection details (host, port, user, password, db) must be manually configured in `config.yaml`.
⚠️For exposing the server publicly (SSE transport), Cloudflare Tunnel (`cloudflared`) must be installed and run separately.

Review RequiredView Analysis

The `run_sql_query` tool, which is central to database interaction, employs a keyword-based filtering mechanism to prevent forbidden operations (delete, drop, update, insert, etc.). This approach is highly vulnerable to bypasses and SQL injection. An attacker could use various techniques such as comments (e.g., `SELECT * FROM users; -- DROP TABLE orders;`), capitalization (`UpDate`), or advanced SQL constructs to circumvent these filters and execute arbitrary malicious SQL commands if the underlying PostgreSQL user account has the necessary permissions. The server's core security objective to prevent data manipulation is not adequately met by the current query validation logic. For instance, `SELECT 1; DROP TABLE users;` would likely bypass the filter. The database connection parameters are specified in `config.yaml`, which should be secured and ideally managed through environment variables or a secrets manager in production, as directly committing it to a repository would expose credentials.

Updated: 2025-12-11GitHub→

Low Cost

VectorForge

by sadatnazarli

Sec8

Provides local, long-term memory for AI assistants (e.g., Claude) using a high-performance C++ vector database.

Setup Requirements

⚠️Requires manual configuration of Claude Desktop's `claude_desktop_config.json` with an absolute path to the server's JavaScript entry point.
⚠️Currently uses mock embeddings; real AI capabilities for embedding generation (e.g., OpenAI, Ollama) are roadmap items and not yet implemented, impacting actual RAG performance.
⚠️Requires C++ (g++ C++17 compatible) and 'make' for building the core C++ vector engine.

Verified SafeView Analysis

The system runs locally and uses standard I/O (stdio) for communication, limiting network attack surface. The Node.js server executes the C++ binary via `child_process.execFile` with explicitly constructed arguments, reducing the risk of command injection through user input. Input parsing in the TypeScript server uses `zod` for validation, and the C++ component parses a simple float array format without 'eval' or similar dangerous functions. No hardcoded secrets were identified. The primary security concern, common in such architectures, is ensuring the C++ binary itself is secure against input-based vulnerabilities, though `strncpy` and fixed-size buffers are used to mitigate some risks.

Updated: 2025-11-24GitHub→

PreviousPage 616 of 760Next