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
berkayinam icon

mcp-server

by berkayinam

Sec9

Provides an HTTP interface for the Model Context Protocol, offering basic tools and compatibility with `mcp-use` and `MCP Inspector`.

Setup Requirements

  • ⚠️Python dependencies (via pip)
Verified SafeView Analysis
The server correctly handles JSON-RPC requests for predefined tools without dynamic code execution. Inputs are processed safely (e.g., `float()` conversion for 'add' tool). CORS is explicitly handled. No critical vulnerabilities like 'eval' or hardcoded secrets were found.
Updated: 2025-11-20GitHub→
0
0
Medium Cost
Sec8

HTTP client for storing and retrieving binary/JSON data via tokens from an MCP Data Buffer server.

Setup Requirements

  • ⚠️Requires a running instance of the MCP Data Buffer server.
  • ⚠️An API token is required for most operations, obtained via `buffer-cli token:create` or similar.
  • ⚠️The `cleanup` method requires admin scope authentication.
Verified SafeView Analysis
This is an HTTP client library. Security risks are primarily related to proper handling of the `apiToken` and ensuring the configured `url` points to a trusted MCP Data Buffer instance. No `eval` or obvious obfuscation detected within the provided context.
Updated: 2025-11-20GitHub→
0
0
Medium Cost
HollowLeaf1981 icon

chessboardmagic-mcp

by HollowLeaf1981

Sec9

Access user's chess repertoires and games, and retrieve TCEC and correspondence chess statistics and games for analysis within AI assistants.

Setup Requirements

  • ⚠️Requires a Chessboard Magic subscription.
  • ⚠️Requires generating a Personal Access Token (PAT) from Chessboard Magic profile.
  • ⚠️Requires Claude Pro and Claude Desktop installed for typical usage.
Verified SafeView Analysis
The server retrieves the Personal Access Token (PAT) from an environment variable, which is good practice. It does not contain 'eval' or other highly dangerous functions. Network requests are made to a specific HTTPS endpoint. Input validation primarily checks for the presence of required arguments (gameId, repertoireId, fen); the backend API is expected to handle deeper validation. There are no obvious injection vulnerabilities identified from the provided source code.
Updated: 2025-12-18GitHub→
0
0
Medium Cost
laica-ayudavets icon

mcp-inmovilla

by laica-ayudavets

Sec8

Enables LLMs to manage properties, clients, owners, and retrieve enumerations from the Inmovilla real estate platform by exposing an MCP-compliant API.

Setup Requirements

  • ⚠️Requires an Inmovilla API Token (obtained from Inmovilla CRM, expires after 3 months of inactivity).
  • ⚠️Requires an OpenAI API Key (a paid service, used by the provided Python testing client).
  • ⚠️Requires a custom 'MCP_API_KEY' for authenticating against the MCP server itself.
  • ⚠️Requires Node.js version 18.19.0 or higher.
Verified SafeView Analysis
The server demonstrates good security practices by using environment variables for API keys and employing Zod for robust input validation across its tools. No 'eval' or obfuscation patterns were found in the provided source code. All external API calls are directed to the legitimate Inmovilla API endpoint. The main security consideration is the broad CORS policy (allowOrigin: "*") which could be a risk if the server is exposed publicly without additional network access controls; however, for a local MCP server, this might be an expected configuration.
Updated: 2025-12-17GitHub→
0
0
High Cost
hyperflow-wms icon

montage-mcp-server

by hyperflow-wms

Sec2

Generates astronomical image mosaic workflows using the Montage toolkit through an MCP server interface, primarily for integration with LLMs like Claude Desktop.

Setup Requirements

  • ⚠️Requires Docker and Docker Compose for execution and testing.
  • ⚠️Requires access to the Docker socket (`/var/run/docker.sock`) for workflow execution.
  • ⚠️Requires network access to the IPAC archive for downloading FITS files.
Review RequiredView Analysis
Critical shell injection vulnerability in workflow generation scripts (`montage-workflow-yaml.py`, `montage-workflow-wfformat.py`): User-controlled inputs (`survey`, `band`) are directly interpolated into shell commands (`mArchiveList`, `mDAGTbls`, `mOverlaps`), allowing arbitrary command execution. Critical container escape vulnerability: The `docker-compose.yml` mounts `/var/run/docker.sock` into the `hyperflow` container, giving it root access to the host's Docker daemon. Use of `yaml.UnsafeLoader` in `server.py` could be a deserialization vulnerability if the generated YAML is ever untrusted, though in current context it loads self-generated content.
Updated: 2025-12-04GitHub→
0
0
Medium Cost

findblock-mcp

by kukapay

Sec8

Provides an MCP server to query EVM blockchain data by timestamp, retrieve chain information, and proxy JSON-RPC calls.

Setup Requirements

  • ⚠️Requires Python 3.10+
  • ⚠️Requires `uv` for dependency management
  • ⚠️Manual configuration of the full installation path for Claude Desktop integration is necessary
Verified SafeView Analysis
The server acts as an RPC proxy, forwarding JSON-RPC calls to external EVM endpoints. The `rpc_call` tool allows arbitrary RPC methods and parameters, which is a powerful feature and relies on the security of the connected RPC nodes and the correctness of the proxy implementation. No explicit use of 'eval' or obfuscation is mentioned, and the code runs locally as an MCP server.
Updated: 2025-11-17GitHub→
0
0
Low Cost
Digital-Defiance icon

mcp-filesystem

by Digital-Defiance

Sec9

Provides advanced filesystem operations for AI agents within strict security boundaries, including batch operations, directory watching, file search/indexing, and permission management.

Setup Requirements

  • ⚠️Requires 'workspaceRoot' to be explicitly configured in a 'mcp-filesystem-config.json' file or via 'MCP_FILESYSTEM_WORKSPACE_ROOT' environment variable.
  • ⚠️If running in Docker directly (not via docker-compose), manual ownership and permission adjustments (e.g., 'chown -R 1001:1001 workspace') for the mounted workspace directory might be necessary.
  • ⚠️The default configuration sets 'requireConfirmation: true' for destructive operations, which may interrupt automated workflows and requires explicit handling or adjustment.
Verified SafeView Analysis
The server implements a robust 10-layer security architecture, including absolute path resolution, workspace boundary checks, path traversal detection, hardcoded system/sensitive path blocklists, configurable allow/block lists, read-only mode, and symlink validation. It features extensive audit logging, rate limiting, and optional user confirmation. The Docker deployment enforces non-root users, minimal capabilities, and network isolation. No direct 'eval' or code obfuscation found; 'eval' is only mentioned in patterns for detecting malicious content, not for execution.
Updated: 2026-01-17GitHub→
0
0
Medium Cost
xxfmin icon
Sec9

A Python server for indexing documents and performing semantic search on them, often used for RAG applications.

Setup Requirements

  • ⚠️Requires Python 3.14 or higher, which is a very recent version and may require environment setup.
  • ⚠️Access to the default embedding model ('Qwen/Qwen3-Embedding-0.6B') or other HuggingFace models may require an `HF_TOKEN` environment variable, which could necessitate a HuggingFace account and potentially incur costs for larger models.
  • ⚠️Relies on `fastmcp` for local development and running, which is a specific framework that users need to install.
  • ⚠️Leverages LanceDB for vector storage, which might have system-level dependencies depending on the operating system.
Verified SafeView Analysis
No direct usage of 'eval' or similar dangerous functions found in the provided Python source code. Sensitive configurations like HuggingFace tokens, server host/port, and CORS origins are properly managed via environment variables. The server exposes HTTP endpoints for communication (SSE/Streamable HTTP), which is standard for web services, but proper CORS configuration is crucial for production deployments. No obvious malicious patterns were detected.
Updated: 2025-12-15GitHub→
0
0
Low Cost
richardichogan icon

azure-schema-mcp

by richardichogan

Sec9

Provides AI assistants with schema discovery for Azure Log Analytics tables and Microsoft Graph API endpoints, enabling accurate KQL query and SDK code generation to eliminate guesswork.

Setup Requirements

  • ⚠️Azure CLI authenticated: `az login`
  • ⚠️Valid `.env` file with AZURE_TENANT_ID and AZURE_WORKSPACE_ID
  • ⚠️Built server: `npm run build`
Verified SafeView Analysis
The server uses DefaultAzureCredential for authentication, a secure and recommended method for Azure services. It explicitly avoids filesystem access for the 'find_working_query_examples' tool due to potential risks, demonstrating responsible design. Local caching of schemas and tokens to specified directories is standard practice. Input validation (Zod) is used for tool arguments. No 'eval' or obvious hardcoded secrets were found.
Updated: 2025-11-26GitHub→
0
0
Low Cost
consigcody94 icon

dev-mirror

by consigcody94

Sec8

Track developer productivity metrics for AI-assisted versus manual coding sessions, providing data-driven insights and comparative reports.

Setup Requirements

  • ⚠️Requires Node.js 18+ to run.
  • ⚠️Requires Claude Desktop to be installed for interaction.
  • ⚠️Manual configuration in Claude Desktop's 'claude_desktop_config.json' file, specifying an absolute path to the server's 'build/index.js' file.
  • ⚠️A full restart of Claude Desktop is necessary after initial configuration.
Verified SafeView Analysis
The server primarily operates on local files within a user's home directory and communicates via stdio. There are no obvious external network calls initiated by the server, nor any direct use of 'eval' or clear command injection vulnerabilities. Data storage is local and user-controlled. The use of fs/promises for file operations in a user's specified data directory is a standard pattern for local applications.
Updated: 2025-11-25GitHub→
0
0
Medium Cost
Ujjwal5200 icon
Sec9

An AI assistant providing specialized tools for math operations and code generation via a Model Context Protocol (MCP) server.

Setup Requirements

  • ⚠️Requires Google Gemini API key (paid or quota-limited)
  • ⚠️Python 3.8+ required
Verified SafeView Analysis
The MCP server code does not contain explicit 'eval', 'exec', or direct system calls. Input validation is present for user queries before passing them to the LLM. Google API keys are loaded securely via environment variables using python-dotenv, preventing hardcoded secrets. The MCP server uses 'stdio' transport for communication, limiting direct network exposure. The main potential risk is the generation of malicious code by the LLM, which the server returns but does not execute. It is the responsibility of the client or end-user to handle generated code securely.
Updated: 2025-12-16GitHub→
0
0
Medium Cost
hiiamhacker23213 icon

french-tax-mcp

by hiiamhacker23213

Sec8

Provides a robust server for French tax calculations and information, designed to assist AI systems with accurate guidance.

Setup Requirements

  • ⚠️Requires Python 3.10 or higher.
  • ⚠️Requires internet access for web scraping official tax authority websites.
  • ⚠️Requires installing Python dependencies including fastmcp, beautifulsoup4, httpx, pydantic, lxml, and markitdown.
Verified SafeView Analysis
The project uses web scraping from official French government websites (impots.gouv.fr, service-public.fr, legifrance.gouv.fr). While the `BaseScraper` implements good practices like rate limiting, caching, and retries with `httpx`, any interaction with external, untrusted content carries inherent risks (e.g., unexpected HTML structure leading to parsing errors, or serving stale data from cache if official sources update). The server does not appear to use `eval` or contain hardcoded secrets. Output is typically structured (JSON, Markdown), reducing the risk of direct script injection if consumed by an AI.
Updated: 2026-01-19GitHub→
PreviousPage 673 of 760Next