Works with the AI Agents you already use
mcp server security audit
Static security auditor for MCP servers to detect injection, confused-deputy, and scope-creep vulnerabilities.
Free
mcp server security audit
Example session with this skill installed
Audit the handlers in src/tools/db_query.ts against their descriptions. Check for SQL injection and ensure tenant_id is derived from the auth token, not the tool arguments.
- Read your context and instructions
- Compiled the mcp server security
Finding: MCP02 (Injection) in src/tools/db_query.ts:42. The 'query' argument is concatenated directly into the raw SQL string. Finding: MCP03 (Confused Deputy) in src/tools/db_query.ts:58. The tenant_id check uses a caller-supplied parameter, allowing cross-tenant access via CWE-639.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
MCP servers often have dangerous drift between tool descriptions and their actual handler code. Developers risk shipping RCE vulnerabilities through unsanitized inputs or leaking tenant data due to confused-deputy patterns in multi-client environments.
What it does
- Performs side-by-side audits of tool docstrings against handler implementations to detect functional drift.
- Traces caller-supplied arguments to identify injection surfaces in shell commands, SQL queries, and file paths.
- Detects confused-deputy vulnerabilities in multi-tenant servers by verifying if authorization checks rely on verified identities or unvalidated tool arguments.
- Identifies over-broad scope declarations and omnibus permissions that violate the principle of least privilege.
- Audits stateful servers for session-handle hijacking risks and shared-mutable-state leaks across concurrent connections.
Frameworks & tools
Model Context Protocol (MCP), OAuth 2.0 (RFC 8707), OWASP MCP Top 10, CVE-2026-54052, CVE-2026-0755, CVE-2026-25536.
Why this beats prompting it yourself
Generic security prompts miss the specific architectural failures unique to the MCP specification, like token passthrough and state-handle binding. This skill enforces a strict multi-step workflow grounded in recent real-world CVEs and primary-source specification requirements.
Use cases
- Reviewing a new MCP server handler for injection risks before its first release.
- Auditing a multi-tenant MCP gateway to ensure tenant isolation isn't bypassable via tool parameters.
- Hardening an existing server by narrowing requested scopes to match actual tool requirements.
- Preparing a formal pre-ship security checklist for third-party review or internal compliance.
Known limitations
Static source review only. It cannot perform dynamic testing against a running server or audit external gateway configurations, environment variables, or CORS policies.
Static source analysis only. Does not perform dynamic adversarial probing or audit external infrastructure like gateways, environment variables, or network configurations.
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
5 uses
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 2 months ago
- Free with an Agensi account
Requires
Permissions
No special permissions declared or detected