PromptBrake Free Tools

Free MCP tools for prompt-injection test inputs, OWASP LLM risk mapping, AI release planning, and custom response test-pack creation for PromptBrake CI.

Hosted MCP Server

npx add-mcp 'https://promptbrake.com/free-tools/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Free Tools for AI Applications: MCP & CI | PromptBrake

Source: https://promptbrake.com/free-tools Summary: Free MCP tools prepare tests for AI applications. The free GitHub Action runs supported checks. Compare both with the licensed Docker runner.

Free Tools

Security testing for AI applications: free tools

Prepare prompt-injection inputs, response and agent tool-call test packs, OWASP risk mappings, and release plans. Free to build and download; no signup required. Web and MCP preparation tools do not scan your application.

MCP, skills, plugins, and GitHub Action setup →

Choose how to test your AI application

Prepare tests for free, run supported checks in GitHub, or use the Docker runner for broader testing and release checks. Test only applications you own or are authorized to test.

Free web and MCP tools: prepare tests

Get prompt-injection inputs, OWASP LLM risk mappings, release plans, and response or tool-call test packs. No PromptBrake account or API key required. These tools do not run scans.

Connect the free MCP tools →

Free GitHub Action: execute supported checks

Run five prompt-injection and five synthetic-secret leakage response checks, or a custom tool-call pack with dispatcher capture. Custom response packs require the licensed runner.

No PromptBrake license required. Model and GitHub runner costs may apply. Captured calls do not verify backend effects or enforce permissions.

Set up the free GitHub Action →

Licensed Docker runner: test, review and retest

Run broader checks and custom response packs in your environment. Review evidence, retest fixes, and apply configurable CI release criteria. License checks and quota usage reach PromptBrake.

Free trial: 3 scans for 7 days, no payment required.

Scan pack: $99 USD once for 5 scans, valid 30 days from payment. No automatic renewal.

Subscription: $499/month for 30 scans and one active runner license. Retests use scan credits; model and infrastructure costs are separate.

See trial and paid runner plans →

Free · No signup

Prepare tests inside your AI assistant

One read-only MCP server works with Claude, ChatGPT, Cursor, and other MCP clients. No PromptBrake account or API key required. Tool inputs and results are not stored; operational request metadata is logged. See our privacy policy.

Server URL https://promptbrake.com/free-tools/mcp

Claude

  1. Open Settings → Connectors
  2. Choose Add custom connector
  3. Paste the server URL

ChatGPT

  1. Open Settings → Connectors → Advanced and turn on Developer mode
  2. Choose Create to add a connector
  3. Paste the server URL and set authentication to None

Claude Code and Cursor

Claude Code, in your terminal:

claude mcp add --transport http promptbrake-free-tools https://promptbrake.com/free-tools/mcp

Cursor: add a remote MCP server with the same URL.

Try this first

After connecting PromptBrake, copy one prompt into your assistant. It may ask a few questions before calling the matching tool.

Prompt-injection test inputs

Use PromptBrake to get direct prompt-injection test inputs for my customer-support chatbot. Explain how to try them on my own staging app; do not run them.

Get fixed adversarial inputs to try on a chatbot you own or are authorized to test. No CI access is needed to generate them; the tool does not run them against your chatbot.

Refund chatbot response checks

Help me build response checks for a refund bot that should ask for human approval.

Get a tests.json pack after choosing the exact response text to check. Creating it is free; running it requires a configured PromptBrake runner and CI access. Text checks do not verify whether a refund was actually authorized.

AI agent release plan

Help me plan my AI agent’s release. Ask me what you need to know.

Get planning gaps and a starter gate policy based on your answers. No CI access is needed to create the plan. It does not execute tests or certify release readiness.

OWASP risk and test ideas

Use PromptBrake to map LLM01 to test ideas for my support chatbot, which reads a knowledge base. Explain what I should inspect in each result.

Get test prompts, warning signals, and suggested ownership. Mapping a risk does not prove your application is vulnerable or protected.

Agent tool-call checks

Use PromptBrake to build a tool-call test pack that checks my agent does not call send_email without approval.

The MCP tool prepares tool-tests.json; it does not execute the test. See the capture setup and examples →

Have your inputs or test pack? Choose how to run your first test →

Give your AI assistant a workflow for our free tools

The PromptBrake skill guides your assistant through prompt-injection inputs, OWASP LLM risk mapping, AI release planning, and custom response test packs.

With Node.js installed, run this in your project and choose your supported assistant:

npx skills add AJ888/promptbrake-skills --skill promptbrake-free-tools

Connect the MCP server above separately. Your client must support both skills and remote MCP. The skill prepares inputs and plans. Run supported checks with the free GitHub Action; custom response packs require a licensed PromptBrake runner. Text checks do not verify backend actions.

View the skill, examples, and manual installation instructions →

AI Test Case Builder

Write your own response checks, download a test file, and run it alongside PromptBrake's security checks in CI.

Build your AI tests →

Agent Tool-Call Checks

Define which tools your agent must avoid calling, or which arguments a tool call must use. Download a test pack for CI.

Requires capture of actual tool calls from your agent. Checks observed invocations, not backend effects or permission enforcement.

Build tool-call checks →

ADLC Release Readiness Planner

Define an AI agent's authority boundaries, human approvals, release blockers, evidence requirements, rollback controls, and starter PromptBrake gate policy.

Build an ADLC release plan →

AI Chatbot Safety Checklist

Review scope, brand promises, human handoff rules, customer data boundaries, and retest discipline before your chatbot goes live.

Open chatbot checklist →

LLM Security Checklist

Use a practical release checklist covering prompt injection, tool permissions, data exposure, output controls, and rerun discipline.

Open checklist →

Prompt Injection Payload Library

Copy direct, indirect, and multi-turn prompt injection examples you can adapt for controlled testing.

Open payload library →

OWASP LLM Test Case Mapper

Translate broad OWASP-style risk categories into concrete test ideas and ownership decisions for your team.

Open test case mapper →

Already have a runner? Set up CI → See sample test results → Need help connecting your application? →