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
trondhindenes icon

code-index-mcp

by trondhindenes

Sec10

Provides fast local source code searching using Zoekt's trigram-based indexing.

Setup Requirements

  • ⚠️Requires Go runtime/toolchain for manual installation or building from source.
  • ⚠️Requires manual configuration (e.g., specifying binary path) in some MCP clients like Claude Desktop if not using the .mcpb bundle.
Verified SafeView Analysis
The server primarily interacts with the local file system for indexing and searching, which is its core functionality. It resolves paths to absolute paths, and includes filtering for common binary and ignored files/directories. The optional web server runs on `127.0.0.1` (localhost), limiting network exposure. There are no indications of 'eval', obfuscation, hardcoded secrets, or malicious patterns. Assuming the user trusts and controls the environment where this binary is run, it appears safe.
Updated: 2025-11-30GitHub→
0
0
Medium Cost
Angelxd0714 icon

mcp-catalog-server

by Angelxd0714

Sec2

Provides a product catalog API, allowing AI agents to interact with product data (search, retrieve, create, update, delete) via the Model Context Protocol (MCP).

Setup Requirements

  • ⚠️Requires a running MongoDB instance.
  • ⚠️Hardcoded MongoDB credentials and connection string: The server attempts to connect to `mongodb://mongo:mongo@localhost:27018/mcp_catalog?authSource=admin` due to hardcoded values in `src/config/db.ts`, ignoring the `MONGO_URI` environment variable specified in `docker-compose.yml`. This requires manual adjustment of the code or ensuring a MongoDB instance is available at `localhost:27018` with these specific credentials.
  • ⚠️Docker-compose MongoDB config mismatch: The `docker-compose.yml` specifies the MongoDB service with port `27017` and hostname `db`, while the application hardcodes `localhost:27018`, leading to connection failure in the provided Docker setup.
Review RequiredView Analysis
The server hardcodes MongoDB credentials ("mongo:mongo") and connection details ("localhost:27018") in `src/config/db.ts`. Crucially, it does not use environment variables (e.g., `process.env.MONGO_URI`) for database connection, even though `docker-compose.yml` attempts to set `MONGO_URI`. This means the application, as written, will always try to connect with the hardcoded, weak credentials and specific host/port, regardless of environment variables, posing a severe vulnerability for any non-local deployment. The `docker-compose.yml` also has a mismatch in MongoDB port (`27017` in compose, `27018` hardcoded in code) and host (`db` in compose, `localhost` hardcoded in code), which would prevent the server from connecting to the database in the provided Docker setup without code modification.
Updated: 2025-11-23GitHub→
0
0
Medium Cost
mimiya525-cloud icon

icon-mcp-server

by mimiya525-cloud

Sec8

Provides developers with fuzzy search for icons from popular libraries (Element Plus, Ant Design) and AI-powered SVG icon generation through a RESTful API and Model Context Protocol (MCP) tools.

Setup Requirements

  • ⚠️Requires at least one AI API Key (e.g., DASHSCOPE_API_KEY, OPENAI_API_KEY) configured in environment variables for AI-powered icon generation. Without suitable keys, AI generation will fail or revert to a very basic fallback.
  • ⚠️Potential Node.js version conflict: 'package.json' specifies '>=14.0.0', but a core dependency ('@modelcontextprotocol/sdk') requires '>=18'. Users might encounter issues if running Node.js versions between 14 and 18.
Verified SafeView Analysis
API keys are loaded from environment variables, which is a good security practice. The server uses standard Express.js and CORS middleware (defaulting to open access). Custom JSON serialization for SVG content and direct embedding of raw SVG into Markdown tables (for MCP responses) means downstream consumers should implement proper sanitization to prevent potential XSS vulnerabilities, although the server itself does not execute user-provided code. No 'eval' or obvious malicious patterns found.
Updated: 2026-01-13GitHub→
0
0
High Cost
HK47196 icon

GhidraMCP

by HK47196

Sec3

GhidraMCP is a Model Context Protocol server that allows LLMs to autonomously reverse engineer applications by exposing numerous core Ghidra functionalities to MCP clients.

Setup Requirements

  • ⚠️Requires a running Ghidra instance with the plugin installed.
  • ⚠️Default server port (8080) might be in use; configurable via Ghidra Tool Options.
  • ⚠️Plugin changes require a Ghidra restart or plugin reload to take effect.
  • ⚠️No built-in authentication/authorization for the HTTP server.
Review RequiredView Analysis
CRITICAL: The server starts an embedded HTTP server (default port 8080) without any visible authentication or authorization mechanisms. This exposes powerful Ghidra functionalities (modifying binaries, setting comments, manipulating structures, querying BSim databases, and arbitrary file imports) to any entity that can reach its IP:PORT. An attacker or compromised LLM client could load and analyze malicious binaries, modify loaded programs, or exfiltrate sensitive program data without user consent. The `importBinary` endpoint allows loading arbitrary files from the filesystem. The `bsim/select_database` endpoint allows connecting to arbitrary BSim databases. While no direct `eval` is present, the exposed APIs have significant side effects and should be heavily protected.
Updated: 2025-12-06GitHub→
0
0
Medium Cost
ufoalan icon
Sec8

Provides Claude with access to Red Hat AMQ 7.12+ via the Jolokia API, enabling message browsing, broker version retrieval, and Keycloak-based authentication/authorization.

Setup Requirements

  • ⚠️Requires Python 3.12+
  • ⚠️Requires a running Red Hat AMQ 7.12.1+ instance accessible from the server
  • ⚠️Requires a running Keycloak server configured with the specified realm, client, and secret
Verified SafeView Analysis
The server retrieves all sensitive configurations (AMQ and Keycloak credentials/endpoints) from environment variables, which is good practice. It leverages Keycloak for robust authentication and UMA-based authorization. AMQ credentials are stored in-memory (not persistent, cleared on logout), but they are stored in plaintext in the `authenticated_sessions` dictionary after Keycloak authentication, which is a minor risk, though necessary for Basic Auth. The communication with Jolokia defaults to HTTP; for production, the README explicitly recommends upgrading to HTTPS via a reverse proxy. No 'eval' or malicious patterns were found.
Updated: 2025-12-01GitHub→
0
0
Medium Cost
Sec9

An enhanced MCP server integrating Granola.ai meeting history and Google Calendar with Claude Desktop, enabling natural language search and unified viewing of meetings.

Setup Requirements

  • ⚠️Requires macOS, Granola.ai, and Claude Desktop installed.
  • ⚠️Requires Python 3.12+ and 'uv' package manager.
  • ⚠️OpenAI API Key is required for intelligent category search (paid service).
Verified SafeView Analysis
Emphasizes 100% local processing for Granola data. Google Calendar uses secure OAuth 2.0 with read-only access. Optional integrations with OpenAI and Turbopuffer send data to those services for advanced features like category search and semantic search, which is explicitly disclosed. No 'eval' or obfuscation is apparent.
Updated: 2025-11-17GitHub→
0
0
Low Cost
alkue-com icon

prompts-mcp

by alkue-com

Sec9

Serves a directory of Markdown files as prompts to any Model Context Protocol (MCP) client for chat sessions.

Setup Requirements

  • ⚠️Requires 'uv' (Ultrafast Python package installer and resolver) to be installed.
  • ⚠️The 'PROMPTS_DIR' environment variable MUST be set and point to an existing directory containing Markdown files, otherwise the server will exit.
  • ⚠️Requires an MCP-compatible client (e.g., Claude Desktop, MCP Inspector, Goose CLI, VS Code/Cursor) to connect and utilize the served prompts.
Verified SafeView Analysis
Uses `subprocess.run` with `shlex.split` and avoids `shell=True`, mitigating command injection risks. `PyYAML` is used with `safe_load`. No obvious hardcoded secrets. The primary external risk is if the `PROMPTS_DIR` environment variable is maliciously configured to point to an untrusted location, allowing arbitrary content to be served.
Updated: 2025-11-20GitHub→
0
0
Medium Cost
patpil-cloudflare-mcp icon

mcp-typescript-docs

by patpil-cloudflare-mcp

Sec9

Provides intelligent semantic search and AI-generated answers over Model Context Protocol (MCP) TypeScript SDK documentation using Cloudflare AutoRAG.

Setup Requirements

  • ⚠️Requires a Cloudflare Workers environment for deployment and execution.
  • ⚠️Requires a WorkOS account for OAuth 2.1 authentication configuration (WORKOS_CLIENT_ID, WORKOS_API_KEY).
  • ⚠️Relies heavily on Cloudflare services: Workers AI (AutoRAG), D1 Database (TOKEN_DB), and KV Namespaces (OAUTH_KV, USER_SESSIONS). These must be configured correctly in Cloudflare Dashboard and `wrangler.jsonc`.
Verified SafeView Analysis
The server demonstrates robust security practices. OAuth 2.1 with PKCE is correctly implemented, preventing authorization code interception. API keys are hashed (SHA-256) before storage and validated against hashes. Database queries use prepared statements to prevent SQL injection. Generated HTML content is escaped to mitigate XSS vulnerabilities. Explicit checks for `is_deleted` status in the database prevent access for revoked accounts. Sensitive API keys (e.g., WORKOS_API_KEY) are loaded from environment variables. AI requests are routed through a Cloudflare AI Gateway, adding a layer for access control and rate limiting. There is no direct evidence of `eval` or similar dynamic code execution being used insecurely.
Updated: 2025-12-28GitHub→
0
0
High Cost
Calq-dev icon

calq-mcp

by Calq-dev

Sec9

Provides a Model Context Protocol server for time tracking, project management, AI-powered memory, and team collaboration for developers, integrating with LLM agents.

Setup Requirements

  • ⚠️Requires paid API keys for Voyage AI (for embeddings) and Anthropic (if using the Discord bot), which will incur token costs based on usage.
  • ⚠️Requires manual setup and configuration of a GitHub OAuth App to obtain Client ID and Client Secret.
  • ⚠️Requires running instances of both PostgreSQL (for structured data) and ChromaDB (for vector embeddings), though Docker Compose streamlines their deployment.
Verified SafeView Analysis
The server uses Drizzle ORM, which mitigates common SQL injection risks through its type-safe API. Authentication is handled via robust OAuth 2.1 with PKCE, leveraging GitHub as the identity provider. Sensitive configurations like API keys and secrets are correctly managed via environment variables. While `POSTGRES_PASSWORD` has a weak default in the `docker-compose.yml`, users are explicitly advised to configure their own. The generated Claude Desktop hook scripts execute Node.js, and perform local file I/O (e.g., reading transcripts); however, the paths are provided by the trusted Claude Desktop environment, minimizing direct arbitrary file access risks from external user input. Overall, the codebase demonstrates good security practices for a modern Node.js application.
Updated: 2025-12-13GitHub→
0
0
Medium Cost
CodingWithCalvin icon

VSC-MCPServer

by CodingWithCalvin

Sec9

Exposes VS Code's language intelligence via a Model Context Protocol (MCP) server for AI coding assistants.

Setup Requirements

  • ⚠️Requires Visual Studio Code (v1.85.0 or later)
  • ⚠️Default port 4000 may conflict with other services; configure `codingwithcalvin.mcp.port` if needed.
  • ⚠️Server binds only to `127.0.0.1` (localhost); remote access is not supported for security reasons.
Verified SafeView Analysis
The server is designed to bind only to localhost (`127.0.0.1`) and includes DNS rebinding protection and CORS restrictions to only allow requests from localhost origins. This significantly limits external attack surfaces. No `eval` or obvious hardcoded secrets were found. The core functionality wraps trusted VS Code API commands, meaning the security model relies on the inherent trust in VS Code's capabilities.
Updated: 2026-01-16GitHub→
0
0
Low Cost
MacroSense-AI icon

macrosense-ai-mcp-server

by MacroSense-AI

Sec4

Provides a template for building Model Context Protocol (MCP) servers with custom tool-calling capabilities for AI models.

Setup Requirements

  • ⚠️Python 3.11 or higher required.
  • ⚠️Docker deployment with `docker-compose.yml` maps server to host port 8080 by default, differing from the local/direct Docker `8000:8000` instructions.
Verified SafeView Analysis
Hardcoded authentication token (`test-token`) in `src/middlewares/auth.py` presents a significant security vulnerability for production deployment. CORS `allow_origins=["*"]` is overly permissive for production. Both should be configured via environment variables or stricter policies.
Updated: 2025-11-29GitHub→
0
0
Medium Cost
farhanrhine icon
Sec9

Analyze Paytm UPI transaction history locally using natural language queries via Claude AI.

Setup Requirements

  • ⚠️Requires Python 3.11+.
  • ⚠️Manual export of Paytm UPI statement as an Excel (.xlsx) file is required.
  • ⚠️Manual conversion of the Excel file to JSON is required before the server can ingest data.
  • ⚠️Requires Claude Desktop to be installed for interaction.
Verified SafeView Analysis
The server demonstrates strong security practices for a local application. It uses parameterized SQL queries to prevent SQL injection, constructs file paths safely to avoid path traversal, and does not execute arbitrary shell commands. All data processing is strictly local with no external network calls for financial data. No hardcoded secrets were found.
Updated: 2025-11-29GitHub→
PreviousPage 745 of 760Next