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
No-Smoke icon

tos-bridge

by No-Smoke

Sec8

Bridges Claude Desktop/Code to a VPS-hosted Token Optimization System using Qdrant (vector database) and Neo4j (knowledge graph) for graph-enhanced search and document management.

Setup Requirements

  • ⚠️Requires a running Ollama server for embedding generation.
  • ⚠️Requires a running Qdrant server for vector storage.
  • ⚠️Requires a running Neo4j server for graph storage and traversal.
  • ⚠️Requires Python 3.10 or newer.
Verified SafeView Analysis
The server uses environment variables for sensitive configurations like Qdrant API keys and Neo4j passwords, which is good practice. Database queries are parameterized, preventing SQL/Cypher injection. There are no obvious signs of 'eval' or direct shell command execution from user input. The primary security consideration for deployment is controlling environment variables (QDRANT_URL, NEO4J_URI, OLLAMA_URL), as malicious manipulation could lead to SSRF (Server-Side Request Forgery) by directing the server to internal network resources. However, this is an operational risk common to services that connect to configurable external endpoints, not a direct vulnerability in the code logic itself.
Updated: 2026-01-16GitHub→
0
0
Medium Cost

MCP-Agent

by jamespani1

Sec5

An AI agent designed to interact with and utilize tools provided by a custom-built MCP (Multi-Component Platform) server.

Setup Requirements

  • ⚠️Requires a custom MCP server instance (likely `mcp_server.py`) to be running before the agent client can connect.
  • ⚠️Dependencies listed in `requirements.txt` must be installed (`pip install -r requirements.txt`).
Review RequiredView Analysis
The project implements a client-server architecture with `mcp_server.py` and `agent_client.py`. This inherently involves network communication. Without access to the source code, it's impossible to verify secure coding practices such as input validation, authentication, authorization, or encryption. This poses potential risks for unauthorized access, data integrity issues, or exposure of server functionalities. No 'eval' or explicit obfuscation methods are apparent from the file list.
Updated: 2025-11-18GitHub→
0
0
Medium Cost
alpha-hack-program icon

cluster-insights-mcp-rs

by alpha-hack-program

Sec9

Provides conversational AI access to Kubernetes cluster resource analysis and capacity planning for platform engineers, SREs, and developers.

Setup Requirements

  • ⚠️Requires Rust 1.70+ and Cargo to build and run locally.
  • ⚠️Requires Kubernetes cluster access with a valid `kubeconfig` configured for read-only RBAC permissions (get/list for nodes, pods, namespaces).
  • ⚠️Requires `jq` for certain development scripts (like version sync), though not for server runtime.
Verified SafeView Analysis
The server is explicitly designed for read-only operations on Kubernetes clusters, which significantly reduces the attack surface compared to tools with write access. It utilizes existing kubeconfig credentials and advises proper RBAC setup. It avoids 'eval' or similar dynamic code execution. The container setup runs as a non-root user (1001) and mentions 'cargo audit' in its CI/CD pipeline, indicating a proactive security stance. The main security consideration is ensuring the provided kubeconfig has appropriately scoped read-only permissions and that the server is not exposed publicly without additional authentication.
Updated: 2025-12-03GitHub→
0
0
Medium Cost
FlyAgileWithGiom icon

miro-remote-mcp

by FlyAgileWithGiom

Sec9

Enables Claude AI to programmatically create and manipulate Miro boards, facilitating complex visualization generation through natural language conversations.

Setup Requirements

  • ⚠️Requires a Miro OAuth application to be created with a Redirect URI configured (e.g., `http://localhost:3003/oauth/callback`) in both Miro settings and the server's configuration.
  • ⚠️Miro OAuth2 tokens (access and refresh tokens) are required and expire, necessitating initial acquisition and potential reauthentication via the `npm run oauth` helper script.
  • ⚠️Requires Node.js (version 16 or newer) and a TypeScript build (`npm run build`) before execution.
  • ⚠️For Claude Desktop/Code, an absolute path to the compiled `dist/index.js` file must be configured, along with client credentials and tokens as environment variables.
Verified SafeView Analysis
The codebase follows good security practices, explicitly avoiding hardcoded secrets by relying on environment variables (often base64 encoded for sensitive keys) or a dedicated token file (`/data/tokens.json` in production). OAuth tokens are managed with automatic refresh and race-condition prevention. Network communication uses `axios` and `https.Agent` for robust connections. No `eval` or obvious malicious patterns found. Security largely depends on proper environment variable management (e.g., in Claude Desktop configuration) and secure file system for `tokens.json`.
Updated: 2026-01-09GitHub→
0
0
Medium Cost
dyelland-lt icon

mcp-example

by dyelland-lt

Sec9

A Model Context Protocol (MCP) server that acts as a bridge between Claude Desktop and the Rick and Morty GraphQL API, demonstrating MCP functionality with HTTPS transport.

Setup Requirements

  • ⚠️Requires Node.js and pnpm to be installed.
  • ⚠️Requires `mkcert` for local SSL certificate generation (with OS-specific installation steps).
  • ⚠️Requires `ngrok` (and a registered authtoken) for public access to connect with Claude Desktop remotely.
  • ⚠️Claude Desktop requires a Pro, Max, Team, or Enterprise plan to use custom connectors.
  • ⚠️Manual SSL certificate generation step `mkcert localhost 127.0.0.1 ::1`.
Verified SafeView Analysis
The server uses HTTPS with locally generated SSL certificates via mkcert, providing TLS security. CORS is broadly enabled (`origin: true`), which is common for development but would ideally be restricted in a production environment. There are no direct uses of `eval`, obfuscation, or hardcoded sensitive API keys (the Rick and Morty API is public). The `start-public.sh` script handles ngrok, which exposes localhost publicly, but this is an explicit action by the user and clearly documented for remote access to Claude Desktop. The `http-server.ts` includes a helpful warning about regenerating SSL certificates for non-localhost IP access, showing security awareness.
Updated: 2025-11-26GitHub→
0
0
Medium Cost

memory-mcp

by mhv2109

Sec9

A Model Context Protocol (MCP) server that provides a SQLite-backed knowledge graph for persistent memory storage to AI clients like Claude.

Setup Requirements

  • ⚠️Requires Go 1.25.1 or later.
  • ⚠️Node.js and npm are required to use the recommended MCP Inspector for interactive testing.
  • ⚠️sqlc is required for generating Go code from SQL queries during development.
Verified SafeView Analysis
The server primarily communicates over stdio (JSON-RPC 2.0), limiting network exposure. It uses a local SQLite database for storage. No obvious 'eval' or obfuscation is present. The primary risk would be malicious input leading to excessive resource consumption or data manipulation within the knowledge graph, but the Go type system and structured schemas mitigate common injection vulnerabilities. Log files are written to /tmp during debug mode.
Updated: 2025-11-18GitHub→
0
0
Low Cost
khushboosaini icon

activity-2-MCP-server

by khushboosaini

Sec8

Demonstrates a minimal server implementation using the `fastmcp` library with a FastAPI fallback for local testing and experimentation.

Setup Requirements

  • ⚠️The `fastmcp` library might require specific versions or API adaptations, as the example includes fallback logic and notes about matching the 'real library API'.
  • ⚠️Relies on Python and `pip` for dependency management, requiring `requirements.txt` installation.
Verified SafeView Analysis
The source code does not contain obvious security vulnerabilities such as 'eval' of user input, hardcoded secrets, or direct system command execution. Endpoints perform simple echo or arithmetic operations. The primary security consideration would be the underlying `fastmcp` or `FastAPI` libraries themselves, which are well-vetted. User input is echoed, which is generally safe but could lead to verbose logs if complex objects are passed.
Updated: 2025-12-11GitHub→
0
0
High Cost
tpitre icon
Sec7

This server acts as a backend for an AI-powered design system documentation and component generation tool, integrating with Storybook to create and modify UI components based on user prompts and design guidelines.

Setup Requirements

  • ⚠️Requires a running Storybook instance on port 6006 (or configured otherwise).
  • ⚠️Requires the Story UI MCP server to be running, typically on port 4005 (or configured otherwise).
  • ⚠️Assumes Caddy is installed and configured with the provided `Caddyfile` for reverse proxying.
  • ⚠️Relies on an external AI provider (e.g., OpenAI, Anthropic, etc.) for code generation, which may incur costs.
Verified SafeView Analysis
The server orchestrates local development tools (Storybook, MCP server, Caddy). While not inherently malicious, several aspects warrant attention: 1. The `/api/raw-source` endpoint in `vite.config.ts` serves local source code with `Access-Control-Allow-Origin: *`, which could expose internal story code if the server is deployed publicly without proper network segmentation. 2. The `StoryUIPanel.tsx` uses a custom markdown renderer and displays AI-generated code. If the AI is prompted to generate malicious code or content (e.g., XSS payloads), this could pose a client-side risk if not adequately sanitized during rendering or if the generated code is executed. 3. Image uploads convert files to base64 and send them to the backend, which is a potential vector if the AI backend processes these images in a vulnerable way. No `eval` or obvious hardcoded secrets were found in the truncated source, and the core scripts use standard command execution.
Updated: 2025-12-02GitHub→
0
0
Low Cost
guthubrx icon

rekall

by guthubrx

Sec8

A local-first, personal knowledge management system for developers, designed to help recall bugs, patterns, and decisions, and integrate with AI assistants via the Model Context Protocol (MCP).

Setup Requirements

  • ⚠️Requires Python 3.10+.
  • ⚠️Semantic search (if enabled) triggers an initial ~90MB model download, requiring internet access.
  • ⚠️Optional 'sqlcipher3-binary' dependency for database encryption.
  • ⚠️Optional 'keyring' dependency for secure passphrase storage.
Verified SafeView Analysis
Rekall is designed as a 100% local, privacy-first tool, storing no API keys or passwords. File permissions for the SQLite database are securely set to 0o600. Outbound network access for semantic embedding model download, source enrichment, and link rot checks is opt-in or clearly indicated. A minor, localized SSRF risk remains in one URL validation path (`validate_url_simple`), although partially mitigated by explicit block patterns. The lack of default encryption for the SQLite database is a design choice for simplicity but is noted as an optional enhancement. Archive import now includes robust checks against zip-bombs and oversized files. CI/CD includes dependency scanning (pip-audit, gitleaks).
Updated: 2025-12-14GitHub→
0
0
Low Cost
JorgeDelAngel03 icon

MCP

by JorgeDelAngel03

Sec9

This server extends LLM capabilities by providing tools for impartial decision-making, weighted randomization, and data persistence of past decisions.

Setup Requirements

  • ⚠️Requires Python 3.10+
  • ⚠️Requires the 'fastmcp' Python library
Verified SafeView Analysis
The server uses `stdio` for communication, limiting direct network exposure. It writes decision history to a local `historial.json` file, which is limited to 50 entries to prevent unbounded growth. There are no obvious hardcoded secrets, `eval` usage, or malicious patterns. The main 'risk' would be sensitive information being stored in the local history file, but it's not directly exposed by the server.
Updated: 2025-11-28GitHub→
0
0
Medium Cost
synergy2test icon

codex-validator-mcp

by synergy2test

Sec8

Validates implementation plans for software development projects using AI, identifying architectural issues, missing dependencies, and best practice violations before code implementation.

Setup Requirements

  • ⚠️Requires Node.js 20+.
  • ⚠️Requires OpenAI Codex CLI (npm install -g @openai/codex) to be installed and authenticated (codex login).
  • ⚠️Requires an OpenAI account with a Pro plan (for Codex CLI) or an OPENAI_API_KEY for fallback API calls.
  • ⚠️Context7 API Key (optional, for higher rate limits on best practices lookups).
Verified SafeView Analysis
The server executes external binaries (OpenAI Codex CLI) and can, in 'apply_changes' mode, modify files. This is an intentional core feature. Safeguards are in place: 'suggest' mode runs in a read-only sandbox by default, and 'apply_changes' mode is designed to prompt for confirmation via the MCP client, though the CLI itself manages the actual application of changes. No hardcoded secrets or arbitrary code execution vulnerabilities were found beyond the intended functionality of an AI coding agent.
Updated: 2026-01-19GitHub→
0
0
Low Cost
KeeVor icon
Sec2

Connects to a MySQL database to perform common operations like listing tables, describing tables, and executing SQL queries.

Setup Requirements

  • ⚠️Requires a running MySQL database instance.
  • ⚠️Requires database connection credentials (DB_HOST, DB_PORT, DB_USER, DB_PASSWORD, DB_NAME) to be set as environment variables.
  • ⚠️Relies on `uv` and `uvx` for installation and execution, which needs to be installed separately.
Review RequiredView Analysis
The `execute_sql` tool allows arbitrary SQL query execution without any apparent input sanitization or restrictions on query types, making it highly vulnerable to SQL injection attacks if exposed to untrusted input. A malicious AI prompt or direct user input could execute destructive commands (e.g., DROP TABLE, DELETE FROM) or exfiltrate data. The database user configured for this server should have the absolute minimum necessary privileges.
Updated: 2025-12-02GitHub→
PreviousPage 612 of 760Next