Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    MCP Security Checklist

    1

    A 12-point, evidence-based security pass over any MCP server before you install it, approve it for a team, or publish your own.

    Free

    Security scanned
    MCP Security Checklist

    MCP Security Checklist

    Example session with this skill installed

    Run a security review on the mcp-server-postgresql repo at commit 8f2a1b. Focus on tool definitions and input validation.

    • Read your context and instructions
    • Compiled the mcp security checklist

    I have completed the security review for mcp-server-postgresql at commit 8f2a1b. The review focused on the manifest and the core query tools.

    MCP security review — mcp-server-postgresql 8f2a1b

    Reviewed: src/index.ts, src/tools/*.ts, package-lock.json · Not reviewed: External database auth provider logic.

    Verdict

    Accept with fixes

    Findings

    SeverityCheckEvidence (file:line or config)ImpactFix
    Critical7src/tools/query.ts:156 uses template literals for table namesSQL injection via model-generated table namesUse a whitelist for allowed tables or use pg-format for identifiers
    High9src/index.ts:45 connects to any provided host stringPotential SSRF if the model provides internal metadata IPsRestrict database hosts to a defined allowlist in environment variables
    Medium10src/tools/read.ts:88 lacks a row limit on resultsA 'SELECT *' on a large table could crash the agent processImplement a hard LIMIT cap (e.g., 1000 rows) and response size truncation
    Low4.env.exampleExample contains a placeholder for a real database URLSmall risk of accidental leak

    Cannot verify

    Check 11 (Supply chain): The postinstall script references a private registry I cannot reach to verify script content.

    Before install

    1. Patch src/tools/query.ts to use parameterized identifiers for table names.
    2. Set PG_ALLOWED_HOSTS in the environment to restrict network reach.

    Next steps

    • Apply the SQL parameterization fix in the tools directory.
    • Update the connection logic to use a host allowlist.
    • Review the postinstall script manually if you have registry access.
    package-lock.json

    Connects securely to your tools. The creator never sees your data.

    About this skill

    MCP servers run with your credentials, on your machine or your network, and often with no human between the model's decision and the tool's side effect. This is the 20-minute pass to run before you install one, approve one for a team, or publish your own.

    What you get

    • A 12-point checklist covering transport exposure, identity, per-tool authorization, secrets, tool-description integrity, destructive tools, input validation, filesystem reach, network reach, timeouts, supply chain and logging.
    • The red flags that end a review on their own, whatever else passes.
    • A report template with severity, evidence (file and line), impact and the smallest fix.
    • An honest "cannot verify" column, because a review that claims full coverage from source alone is not true.

    How it works

    Point the agent at a repository and a commit. It reads the manifest, tool definitions and transport setup, then fills in the table. Every finding cites a file and line, or an exact config value. Nothing is executed and no secrets are requested.

    Who it is for

    Anyone about to wire a third-party MCP server into an agent with real access: local setups, team rollouts, and publishers doing a last check before release.

    Limits

    Configuration and code shape only. It does not test a live endpoint and does not replace a dependency advisory audit.

    Free starter. For the full audit - 60 checks, tool-description poisoning, confused-deputy flows, multi-tenant isolation and OAuth scope design - see MCP Server Auditor: https://www.agensi.io/skills/mcp-server-auditor

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Requires access to source code for full verification. Does not perform live endpoint testing or deep dependency vulnerability scanning.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    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

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 2 days ago

    • Free with an Agensi account

    Software
    Auditor MCP

    Permissions

    Read Files
    Listed2 days ago

    What's inside

    Frequently Asked Questions

    This skill performs an automated, 12-point technical audit of an MCP server's source code and manifest to identify critical risks like prompt injection, path traversal, and insecure credential handling before you grant the server access to your system.

    The skill is compatible with any AI agent that supports MCP and can analyze repositories written in TypeScript, JavaScript, or Python.

    No, this skill performs a static analysis of the repository and manifest; it does not require the target MCP server to be running or installed, making it safe to use for pre-installation vetting.

    You will receive a pass/fail report for each of the 12 security checkpoints, complete with specific file and line-number evidence for any identified vulnerabilities or red flags.

    While it provides deep evidence-based analysis of code and configuration, it does not perform live endpoint testing or a full audit of third-party dependencies (supply chain vulnerabilities).

    New in LLM Ops

    Free