Skip to main content
Home
AI Agentic Workflows

AI covers the breadth. Your testers cover the depth.

Trigger your own AI agents from any AttackForge project. They test, retest, enrich findings and draft summaries through Flows and 80+ MCP tools, and every result lands back in the project for your team to review and sign off.

Start Free Trial
  • Bring your own agents
  • Per-user tool grants
  • Human sign-off
  • Run history
How it works

One click starts the agent. Your process stays in charge.

  1. 1

    Trigger

    A user runs an Action such as "Launch AI Hackbots" on a project. An optional form on the Action becomes the agent's prompt for that run.

  2. 2

    Flow

    An AttackForge Flow calls your agent over HTTP with the project and the form input. The agent then reads the project's scope and assigned test cases itself, over MCP.

  3. 3

    Your agent

    Runs on your platform, such as Microsoft Copilot Studio or OpenClaw with Microsoft Foundry. AttackForge orchestrates the work; it does not host the model.

  4. 4

    Results over MCP

    The agent writes vulnerabilities, test case results, remediation notes and summaries back into the project, using only the MCP tools it has been granted.

  5. 5

    Your team reviews

    Testers verify, QA and sign off, exactly as they would for human findings.

If you can trigger it over HTTP and connect it over MCP, it can work inside AttackForge.

Five agentic workflows

Agentic penetration testing

Launch your agents from a project. They find and enrich vulnerabilities and stream them into the project as they go, for your testers to verify.

What lands in AttackForgeVulnerabilities, evidence, test case results.

Agentic retesting

When findings are Ready for Retesting, agents re-test them, re-open or close each one and write remediation notes, and your team confirms every result.

What lands in AttackForgeRetest status and remediation notes.

Agentic vulnerability enrichment

Agents research CVE and NVD, EPSS, KEV and Exploit-DB, map findings to MITRE ATT&CK, CWE, CAPEC, OWASP, PCI-DSS, ISO 27001 and NIST, and write the results back into your writeups for your reviewers to check.

What lands in AttackForgeEnriched writeups, ready for review.

Agentic executive summaries

Agents draft the executive summary straight into the project's Reporting fields, ready for review before you generate the report.

What lands in AttackForgeReporting fields.

Agentic attack-chain analysis

Agents analyse how a project's findings chain together and upload the analysis to the project Workspace.

Pair it with Automated AI Attack Chains: an Interactive Attack Chain Explorer, MITRE ATT&CK and Cyber Kill Chain views built from real findings with AttackForge Skills (available on request).

What lands in AttackForgeWorkspace files.
Trust and control

Built for teams that have to answer for every finding.

Per-user tool grants

Every MCP tool is off until an administrator grants it, per user, per group or through the API. Connect each agent as its own user and grant it only the tools its workflow needs.

  • Bring your own agents

    Your models, your platform. Project data goes to the agents and MCP clients you connect, so the data boundary is the one you choose. AttackForge triggers the work and receives the results.

  • Controlled triggers

    Share Actions and Flows with specific roles, groups or users, with trigger rights set separately. Effective Access shows exactly who can run each one.

  • Human sign-off

    Agents cover the breadth; your testers cover the depth. Agent output goes through your normal review and QA before it reaches a report, and with new vulnerabilities set to Pending, only users with Edit access see them until they are released.

  • Run history

    Each Action run shows which Flows it triggered and their status, and Flow run logs record the request and the input and output of every step.

Methodology first

Agents follow your methodology, not their own.

Give agents a bounded plan: they work strictly through the test cases assigned to the project, from OWASP WSTG, ASVS and MASTG to MITRE ATT&CK or your own suites. A linked vulnerability fails the test case, so coverage and evidence line up for every reviewer.

Prefer exploration? Let agents record their own test procedures as new project test cases, so every step they took leaves an evidence trail.

By the numbers

Agents can create and update vulnerabilities, writeups, test cases, remediation notes, reporting fields and more, within the tools you grant.

  • 80+

    MCP tools, read and write

  • Unlimited

    agentic workflows

  • Zero

    MCP tools on until an administrator grants tools access

Works with remote and local models.

Any agent you can trigger over HTTP and connect over MCP.

Agent platforms (reference stacks)

  • Microsoft Copilot Studio

  • OpenClaw with Microsoft Foundry

  • Custom Harness

MCP clients

  • Claude

    Remote

  • ChatGPT

    Remote

  • Microsoft Copilot Studio

    Remote

  • Claude Desktop

    Local, via the @attackforge/mcp-server package

  • LM Studio

    Local, via the @attackforge/mcp-server package

Frequently Asked Questions

Put your agents to work inside your methodology.

Start Your Free Trial
  • SOC 2 Type II Certified
  • Instant Deployment
  • No credit card required