Works with the AI Agents you already use
MCP Server Security
Professional security auditing for MCP servers and configurations to prevent injection, data leaks, and tool misuse.
$20
MCP Server Security
Example session with this skill installed
**Audit this MCP config:** {"mcpServers": {"weather": {"command": "npx", "args": ["-y", "@modelcontextprotocol/server-weather"], "env": {"API_KEY": "sk-12345"}}}}
- Read your context and instructions
- Compiled the mcp server security
The audit of the provided MCP configuration is complete. One high-severity finding and one medium-severity finding were identified regarding credential handling and package integrity. ## MCP Security Audit Report **Verdict:** Do not deploy until fixed **Security Posture Score:** 45/100 (Grade: F) ### Executive Summary The configuration contains a plaintext API key in the environment variables and uses an unpinned `npx` command. These flaws allow for credential theft and potential supply-chain attacks if the package version is compromised. ### Findings | ID | Finding | Category | Severity | Confidence | |:---|:---|:---|:---|:---| | MCP-06 | Plaintext Secret in Config | Auth Model | High | Confirmed | | MCP-09 | Unpinned Remote Package | Supply Chain | Medium | Confirmed | #### [MCP-06] Plaintext Secret in Config - **Component:** `mcpServers.weather.env.API_KEY` - **Description:** The `API_KEY` is stored as a raw string in the JSON configuration file. - **Risk:** Any process or user with read access to the config file can steal the credential. - **Remediation:** Use an environment variable reference or a secret manager. For Claude Desktop, set the variable in the shell environment instead of the JSON file. #### [MCP-09] Unpinned Remote Package - **Component:** `command: npx` - **Description:** The use of `npx -y` without a version tag downloads the latest version of the server every time it runs. - **Risk:** A "rug pull" or package hijack could result in malicious code execution during the next startup. - **Remediation:** Pin the version explicitly: `npx -y @modelcontextprotocol/server-weather@1.0.2`. **Next steps** 1. Move `API_KEY` to your local environment variables and remove it from the JSON. 2. Update the `args` to include a specific version tag for the weather server. 3. Re-run the audit once the configuration is updated.
Connects securely to your tools. The creator never sees your data.
About this skill
Security debt in Model Context Protocol (MCP) implementations can lead to full system compromise through tool poisoning or command injection. This skill provides a systematic security audit of MCP servers, client configurations, and agent deployments to identify vulnerabilities before they are exploited. ## What it does - **Vulnerability detection** identifies common MCP threats like tool shadowing, prompt injection via tool descriptions, and confused deputy attacks. - **Static analysis** reviews server source code and `claude_desktop_config.json` files for risky patterns, unpinned packages, and plaintext secrets. - **Attack surface mapping** inventories tools, resources, and trust boundaries to calculate the blast radius of potential exploits. - **Security scoring** generates a posture score from 0-100 based on a standardized severity rubric to guide deployment decisions. - **Remediation planning** provides actionable, code-level fixes and hardening steps for every identified finding. ## How it works 1. **Scope definition** establishes the target type, transport method (stdio/SSE), and trust context of the MCP deployment. 2. **Threat cataloging** maps the implementation against 18 specific MCP threat categories (MCP-01 to MCP-18). 3. **Evidence collection** traces tool arguments from schema to sinks like shell execution or filesystem access. 4. **Report generation** delivers an executive summary, findings list, and a prioritized remediation roadmap. ## Frameworks & tools This skill is designed for the Model Context Protocol ecosystem, covering Python and Node.js server implementations, JSON configurations, and clients like Claude Desktop, Cursor, and VS Code. ## Why this beats prompting it yourself Generic security prompts often miss MCP-specific vectors like tool description injection or rug-pull updates. This skill uses a specialized threat catalog and a multi-phase audit workflow to ensure no trust boundary is overlooked. ## Use cases - **Pre-install vetting** to decide if a third-party MCP server is safe for a local machine. - **Developer self-review** to secure custom tools before distributing them to a team. - **Config hardening** to audit `mcpServers` settings for over-privileged filesystem access. - **Threat modeling** to visualize trust boundaries between the user, model, and server. ## Known limitations Static analysis cannot guarantee 100% detection of runtime logic flaws. Audits are limited to the code and configuration provided or described by the user.
Static analysis cannot detect all runtime logic flaws. Results depend on the completeness of provided source code or configuration files. Verification of remote servers requires explicit authorization.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Requires
Permissions
Allowed Hosts