Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    mcp gateway config auditor

    1

    Independent configuration audit for MCP gateways to verify auth, rate-limiting, and multi-tenant isolation.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    mcp gateway config auditor

    mcp gateway config auditor

    Example session with this skill installed

    Audit my Kuadrant MCP gateway config. I've attached the AuthPolicy, the rate-limit CRD, and a sample of our JSON logs. We need to ensure Tenant A cannot exhaust Tenant B's tool budget.

    • Read your context and instructions
    • Compiled the mcp gateway config

    I have audited the Kuadrant configuration artifacts provided. The review focused on auth logic, rate-limit scoping, and tenant isolation provenance.

    MCP Gateway Audit Findings

    SeverityFindingImpactIron Rule Violation
    HighToken Passthrough (Merged)Bypasses per-tenant rate limits and breaks log attribution.Rule 5, Rule 2
    HighIsolation TheatreTenant ID is read from a header (x-tenant-id) rather than a verified JWT claim.Rule 3
    MediumGlobal Rate-Limit OnlyRateLimitPolicy lacks a domain key for per-tenant budgeting.Rule 4
    LowLog Identity OpacityLogs record the Gateway service account GUID instead of the resolved user UPN.Rule 6

    Analysis

    The root issue is the AuthPolicy configuration on the /v1/tools route. It validates the JWT signature but forwards the raw token to downstream servers without extracting and injecting the tenant_id into the context. Consequently, the rate-limiter sees aggregate traffic instead of individual tenant streams. Additionally, the admin UI at /admin/stats has no AuthPolicy attached, leaving gateway metrics exposed to the public internet.

    Next steps

    • Update the AuthPolicy to use CEL expressions that verify the role claim.
    • Map the tenant_id rate-limit descriptor to the auth.claims.tenant metadata.
    • Apply an explicit AUTH_REQUIRED policy to the /admin and /metrics routes.

    Corrected Configuration (Remediation)

    
    # Fixed RateLimitPolicy for per-tenant scoping
    apiVersion: kuadrant.io/v1beta2
    kind: RateLimitPolicy
    metadata:
      name: mcp-tenant-limits
    spec:
      targetRef:
        group: gateway.networking.k8s.io
        kind: HTTPRoute
        name: mcp-tools-route
      limits:
        "per-tenant-quota":
          rates:
            - limit: 100
              window: 1m
          counters:
            - metadata:
                - key: auth.claims.tenant # Pulls from verified token, not header
    

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

    About this skill

    MCP gateways often suffer from "theatre of security" where policies exist but enforce nothing. An AuthPolicy that checks identity but ignores roles, or a tenant isolation check that trusts a caller-supplied header, leaves your infrastructure exposed. This skill provides a static configuration audit for deployed MCP proxies like Kuadrant, IBM mcp-context-forge, Kong, or Traefik Hub to find these gaps before they are exploited.

    What it does

    • Route Enumeration identifies every exposed endpoint, including often-missed admin UIs, metrics, and documentation paths.
    • Policy Logic Audit reads actual CEL expressions and rules to ensure they enforce specific roles, not just presence of identity.
    • Isolation Trace follows tenant identifiers to their source to verify they originate from trusted tokens rather than request arguments.
    • Rate-Limit Scoping validates whether limits apply per-tenant, per-tool, or globally to prevent resource exhaustion.
    • Log Attribution Review samples log schemas to ensure resolved caller identities are recorded instead of opaque gateway credentials.

    How it works

    1. Establish Scope by listing available policy files, environment variables, and log samples for the specific gateway product.
    2. Execute Audit across auth, rate-limiting, isolation, and logging surfaces using the MCP specification checklist.
    3. Cross-Check Findings to merge related issues, such as token passthrough defeating both rate limits and logging simultaneously.
    4. Generate Remediation (paid tier) with corrected configuration patches and a pre-deployment sign-off checklist.

    Frameworks & tools

    Works with any MCP gateway or AI proxy supporting policy-based configuration, including Kuadrant, IBM mcp-context-forge, Kong AI Gateway, and Traefik Hub.

    Why this beats prompting it yourself

    Generic prompts often miss the "shared root cause" problem where one configuration flaw invalidates multiple security layers. This skill uses a specialized seven-layer isolation model and a specific 2026 incident database to catch combined vulnerabilities like CORS wildcarding paired with unauthenticated passthrough endpoints.

    Use cases

    • Reviewing gateway configuration before a production go-live to ensure multi-tenant isolation.
    • Auditing a deployment after an identity provider change to verify token claims are still mapped correctly.
    • Troubleshooting why logs cannot attribute specific tool calls to individual users or tenants.

    Known limitations

    Does not perform live adversarial testing or dynamic probing. Does not audit the source code of the underlying MCP servers or recommend switching between gateway products.

    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.

    Static config review only. No live traffic probing. Does not audit MCP server code or tool handler logic. Limited to auth, rate-limiting, isolation, and logging surfaces.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Software
    Context Forge MCPwhen a buyer has already deployed an MCPSecurity Audit MCPGateway MCPAudit Gateway MCPAuth Proxy MCPAudit MCP

    Permissions

    No special permissions declared or detected

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    The auditor identifies 'theatre of security' flaws, such as policies that verify identity but fail to enforce roles, or tenant isolation that relies on untrusted request headers instead of secure tokens.

    It is built around Kuadrant's MCP gateway and IBM mcp-context-forge, and also covers other MCP gateways and proxies such as Kong AI Gateway and Traefik Hub on the same checklist (auth, rate limits, tenant isolation, logging). Coverage is deepest for the two named first. Any gateway whose policy files, environment variables and logs you can share can be reviewed against it.

    No, this is a static configuration audit tool; it analyzes your policy files, environment variables, and log schemas without performing live penetration testing or active probing.

    A ranked list of findings across auth, rate-limit scoping, multi-tenant isolation and logging. Each finding names the file or setting reviewed, the spec section or documented misconfiguration it relates to, and what could not be checked from the files provided. The skill also generates corrected configuration (policy rules, route-level auth, tenant-key derivation, CORS allowlist, log-schema fields) and a pre-sign-off checklist.

    Your gateway's policy or CRD files and environment variables, plus CORS settings. Log samples are needed to check logging completeness; redact personal data before sharing them. If something is missing, the report says which checks it could not complete.

    It follows a fixed checklist grounded in the MCP specification, documented gateway defaults, a seven-layer multi-tenant isolation model, and a dated 2026 incident. It treats combined problems (for example a wildcard CORS policy together with an unauthenticated endpoint) as one higher-severity finding rather than two low ones. It also states what it could not verify, instead of saying "looks fine".

    New in LLM Ops