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
sunny-suzega icon
Sec1

This repository is intended to be an AI application onboarding solution, but its specific functionality and implementation details cannot be determined as the source code was not provided for analysis.

Setup Requirements

  • ⚠️Source code is missing, thus specific setup requirements, dependencies, or potential friction points (e.g., API keys, specific software versions, Docker) cannot be identified.
  • ⚠️A truncated README was mentioned, but no content was provided, limiting context.
Review RequiredView Analysis
CRITICAL: No source code was provided for analysis. Therefore, a comprehensive security audit is impossible. The low score reflects the inability to verify the project's safety against malicious patterns, 'eval' usage, obfuscation, hardcoded secrets, or network risks.
Updated: 2025-12-11GitHub→
0
0
Low Cost
Sec4

A serverless Cloudflare Worker acting as a remote endpoint or proxy, potentially for a Minecraft-related service, designed to operate without built-in authentication.

Setup Requirements

  • ⚠️Requires a Cloudflare account for deployment.
  • ⚠️Requires `wrangler` CLI to be installed and configured.
  • ⚠️Operates without built-in authentication, necessitating external security considerations for public deployment.
Verified SafeView Analysis
The server is explicitly designed as 'authless', meaning it lacks intrinsic authentication. If deployed publicly without external security measures (e.g., Cloudflare Access, IP restrictions), it presents a significant network risk, allowing anyone to access its endpoints. No obvious signs of 'eval' or obfuscation from the file list.
Updated: 2025-11-18GitHub→
0
0
Medium Cost
ABINETHACKER icon

MUSTer_MCP

by ABINETHACKER

Sec8

Automates interaction with Macau University of Science and Technology (M.U.S.T.) campus systems (Moodle, Wemust) to facilitate tasks like schedule retrieval, course content access, and resource downloads for LLM integration.

Setup Requirements

  • ⚠️Requires Python 3.12+.
  • ⚠️Requires a local installation of Chrome browser and its corresponding Chromedriver for Selenium to function.
  • ⚠️Installation instructions recommend using `uv` (a fast Python package manager).
  • ⚠️Mandatory environment variables: `MUSTER_USERNAME`, `MUSTER_PASSWORD`.
Verified SafeView Analysis
The server correctly uses environment variables for sensitive credentials (username/password). It does not use `eval` or exhibit overt malicious patterns. However, it relies on Selenium for browser automation, which can introduce risks: 1. The `--no-sandbox` Chrome option is used, which can be a security concern in certain multi-user or untrusted environments. 2. The `download_resource` and `open_URL_with_authorization` tools accept `resource_url` and `url` parameters respectively. While intended for Moodle/Wemust, a malicious LLM prompt could potentially direct these tools to interact with arbitrary, untrusted URLs or download files from them, leading to unexpected behavior or system exposure. Users should ensure LLM interactions are guarded against such misuse.
Updated: 2026-01-19GitHub→
0
0
Low Cost
VeVarunSharma icon
Sec1

Demonstrates security vulnerabilities and mitigations in Model Context Protocol (MCP) servers for educational purposes.

Setup Requirements

  • ⚠️Contains intentionally vulnerable code: DO NOT use the vulnerable examples in production systems.
  • ⚠️Requires Node.js 18+ and TypeScript 5.3+ to run.
Review RequiredView Analysis
This repository is intentionally designed to showcase critical security vulnerabilities (Command Injection, Path Traversal, SSRF, Tool Poisoning, Full-Schema Poisoning, Advanced Tool Poisoning) in its 'vulnerable' implementations. Running the vulnerable server code, or any part of it without careful application of the provided 'secure' mitigations, would lead to severe compromises including arbitrary code execution, sensitive data exfiltration (e.g., SSH keys, AWS credentials), and internal network exposure. While the repository provides secure examples and strong warnings, its core content for demonstration is highly insecure by design. Therefore, it is extremely unsafe to run in any non-isolated or production environment.
Updated: 2025-11-27GitHub→
0
0
Low Cost
PROGRAMER122223 icon

MCP-server

by PROGRAMER122223

Sec9

Provides a Model Context Protocol (MCP) server for basic arithmetic operations (add, subtract, multiply, divide).

Setup Requirements

  • ⚠️Requires Windows Package Manager (winget) for initial setup steps.
  • ⚠️Requires NVM for Windows and Node.js for the MCP Model Inspector and possibly the `mcp` CLI tools.
  • ⚠️Requires Python 3.13+ for running the server.
Verified SafeView Analysis
The server uses basic arithmetic operations and local file reads (for the `swiggy-mcp1` part). No direct use of `eval()` or `exec()` is observed. The file reading in `swiggy-mcp1/customercare.py` uses fixed filenames or a parameter that is not used to construct the file path, mitigating direct path traversal risks. The `calc_mcp/calc.py` server does not involve file I/O or external network calls beyond the MCP protocol itself. Overall, the direct source code provided appears to be reasonably secure for its stated purpose.
Updated: 2025-12-15GitHub→
0
0
Medium Cost
omniagentpay icon

mcp-server

by omniagentpay

Sec5

Serves as a production-ready FastMCP server for AI agents to securely manage wallets and execute payments via OmniAgentPay payment tools.

Setup Requirements

  • ⚠️Requires Python 3.11+.
  • ⚠️Requires configuration with OmniAgentPay payment credentials (CIRCLE_API_KEY, ENTITY_SECRET), which are validated and required in 'prod' environment.
  • ⚠️Authentication (MCP_AUTH_TOKEN or MCP_JWT_SECRET) is enabled and required by default; local setup requires generating a token and setting it in `.env` or as an environment variable.
Verified SafeView Analysis
The core application uses FastAPI with Pydantic for input validation, supports Bearer token/JWT authentication, and integrates payment guardrails (budget, rate limit, recipient whitelist) via OmniAgentPay SDK. Structured logging and a health check are present. However, hardcoded example secrets (API keys, auth tokens) are directly embedded in the `deploy-first-time.sh` and `cloudbuild.yaml` scripts as environment variables. While documentation advises using secure methods like Google Cloud Secret Manager, the provided deployment scripts themselves demonstrate a critical security anti-pattern by exposing secrets in plaintext, which could lead to accidental exposure if real keys are used similarly. Webhook signature verification is bypassed in 'dev' environment and requires `OMNIAGENTPAY_WEBHOOK_SECRET` for production.
Updated: 2026-01-18GitHub→
0
0
High Cost
hao-hoang icon
Sec8

An autonomous trading floor application with multiple AI agents making trades and real-time portfolio monitoring.

Setup Requirements

  • ⚠️Requires OpenAI API Key (Paid)
  • ⚠️Requires Polygon API Key (Free for EOD, Paid for Real-time)
  • ⚠️Requires Node.js and npm for some MCP servers (Fetch, Brave Search, Memory)
  • ⚠️Docker and Docker Compose are highly recommended for deployment
  • ⚠️Specific Python versions required (3.10, 3.11, or 3.12)
Verified SafeView Analysis
The application relies heavily on external API keys (OpenAI, Polygon, Brave Search, DeepSeek, Google, Grok, OpenRouter, Pushover) which are loaded from environment variables (.env file), following good security practices for secrets management. Inter-process communication for MCP servers largely uses standard I/O (stdio_client) which is secure for local communication. However, some MCP servers are sourced and executed via 'npx' or 'uvx --from git+', which involves downloading and running external packages at runtime. While these are common package management patterns, they introduce a supply chain risk if the remote repositories were compromised. The Streamlit UI is exposed on port 8501, which is standard for web UIs.
Updated: 2026-01-17GitHub→
0
0
Low Cost
varunshah3129 icon

yahoo-finance-mcp-server

by varunshah3129

Sec6

Integrate Yahoo Finance data and analytics into AI applications and provide an interactive dashboard for financial queries.

Setup Requirements

  • ⚠️Requires Node.js 18.0.0 or higher.
  • ⚠️Requires a local Ollama server running with the `phi3:mini` model downloaded and accessible at `http://localhost:11434`.
  • ⚠️The internal `yahoo-finance-mcp` server module must be built (`cd yahoo-finance-mcp && npm install && npm run build`) before running the main HTTP bridge.
  • ⚠️For LibreChat integration, the `cwd` path in `librechat-config.json` needs to be manually updated to the server's directory.
Verified SafeView Analysis
The client-side React application uses `dangerouslySetInnerHTML` for rendering markdown content from the MCP server, which can be an XSS risk if the MCP server is compromised or produces malicious output. The HTTP bridge uses `cors()` without specific origin restrictions, posing a network risk if exposed externally. Reliance on a local Ollama server means its security and integrity are paramount. These risks are somewhat mitigated by the expected local/self-hosted deployment within a trusted environment.
Updated: 2025-11-26GitHub→
0
0
Medium Cost
GSNadarajan icon

figma-mcp-server

by GSNadarajan

Sec6

Provides a Model Context Protocol (MCP) server to expose Figma design tools to Claude Desktop for design-to-code workflows.

Setup Requirements

  • ⚠️Requires a Figma Personal Access Token for API authentication.
  • ⚠️Claude Code's MCP server configuration is project-specific, requiring careful setup or CLI usage per project.
  • ⚠️For production use, requires deployment to a hosting service like Render, Railway, or Fly.io.
Verified SafeView Analysis
The server uses `CORSMiddleware` with `allow_origins=["*"]`, which permits requests from any origin. While common for specific integrations, it can be overly permissive for general web services. Additionally, the `/save-code` endpoint allows writing arbitrary HTML, CSS, and JavaScript strings to the server's filesystem. If this endpoint is publicly exposed without robust authentication and authorization, it could be exploited to store malicious files (e.g., web shells, spam), consume disk space, or potentially lead to other vulnerabilities if the saved files are later served or processed by other systems.
Updated: 2025-11-25GitHub→
0
0
Medium Cost
BERDataLakehouse icon

datalake-mcp-server

by BERDataLakehouse

Sec7

A FastAPI service enabling AI assistants to interact with Delta Lake tables stored in MinIO through Spark, using the Model Context Protocol (MCP) for natural language data operations.

Setup Requirements

  • ⚠️Requires KBase authentication token and 'BERDL_USER' role for access.
  • ⚠️Initial login to BERDL JupyterHub is mandatory for user provisioning (MinIO buckets, permissions, namespace isolation).
  • ⚠️MinIO credentials must be stored in `/home/{username}/.berdl_minio_credentials` for each user.
  • ⚠️Python 3.13+ is required (from `pyproject.toml`).
  • ⚠️Building from source requires cloning `spark_notebook_base`, `kube_spark_manager_image`, and `hive_metastore` repositories at the same directory level.
Review RequiredView Analysis
The service explicitly warns that it allows arbitrary 'read-oriented' queries to be executed against Delta Lake tables and that query results will be sent to the model host server (unless locally hosted). It also strictly warns against deploying to production without KBase leadership approval and advises against querying sensitive data. Positive security aspects include: - KBase authentication with role-based access control (`KBASE_REQUIRED_ROLES`). - Strong input validation for structured SELECT queries to prevent SQL injection (`_validate_identifier`, `_escape_value`). - A basic (but not comprehensive) `_check_query_is_valid` function to disallow common destructive SQL keywords (`drop`, `delete`, `insert`, `update`, `create`, `alter`, `merge`, `truncate`, `vacuum`), metacharacters (`;`, `/*`, `--`), and PostgreSQL system schemas (`pg_`, `information_schema`). - Dynamic loading of MinIO credentials from user home directories (`/home/{username}/.berdl_minio_credentials`), avoiding hardcoded secrets for data access. - Explicit validation in `src/main.py` requiring `POSTGRES_USER` to be `readonly_user`, indicating a read-only database connection for the Hive Metastore backend. - Use of LRUCache for KBase Auth tokens and Redis for query results, with user-scoped cache keys (`_generate_cache_key`) to prevent cross-user data leakage from the cache. Areas for caution/improvement: - The `_check_query_is_valid` function is not comprehensive and explicitly noted as such. Reliance on it for complete SQL injection prevention is not advised. The primary defense for arbitrary `read-oriented` queries is the read-only database user and S3 policies. - The core risk remains the 'arbitrary read-oriented queries' which, if misused by an AI or malicious actor, could expose data that the user *shouldn't* have access to, even if read-only, if S3 policies aren't perfectly aligned with the KBase Auth roles/groups logic. - The warning about sending query results to the model host server highlights a significant data privacy concern for users.
Updated: 2026-01-17GitHub→
0
0
High Cost
yigitkonur icon

latitude-mcp-server

by yigitkonur

Sec8

Manages PromptL prompts for Latitude.so, enabling AI-powered prompt lifecycle management including validation, versioning, and execution directly from MCP clients.

Setup Requirements

  • ⚠️Requires Latitude.so API Key (paid service to execute prompts)
  • ⚠️Requires Latitude.so Project ID
  • ⚠️Requires Node.js 18.0.0 or higher
Verified SafeView Analysis
Uses environment variables for API keys (good practice). Client-side input validation is performed using the 'promptl-ai' library to prevent invalid PromptL syntax. File system operations are scoped to intended functionality (reading/writing .promptl files, creating temporary directories for sync operations). The primary security concerns relate to securing the LATITUDE_API_KEY from unauthorized access and the upstream Latitude.so API's handling of prompt execution.
Updated: 2026-01-19GitHub→
0
0
Medium Cost
Sri-Manikandan icon

mcp-docs-server

by Sri-Manikandan

Sec6

A Streamlit and FastAPI-based client for an AI agent that uses an LLM to search various technical documentation platforms via web search.

Setup Requirements

  • ⚠️Requires Anthropic API Key (Paid service)
  • ⚠️Requires Serper API Key (Paid service for web search)
  • ⚠️Requires Python 3.12+
Verified SafeView Analysis
The FastAPI application uses `CORSMiddleware` with `allow_origins=["*"]`, which is a significant security risk for any public-facing API as it permits requests from any domain. The `server_script_path` used to launch the MCP server is configurable via environment variables; if an attacker can control this path, it could lead to arbitrary code execution. No direct `eval` or clear obfuscation found. External HTTP requests are made to Serper API and arbitrary URLs for documentation fetching, which are controlled by the tool's logic but involve external services.
Updated: 2025-12-15GitHub→
PreviousPage 735 of 760Next