Concepts

How Dapper works under the hood. These pages explain the design that lets an LLM behave like a human penetration tester: a four-phase methodology, a team of specialist agents running on a durable workflow engine, and the tooling that lets those agents log in, drive a real browser, and prove a finding by exploiting it.

If you’ve only read the Quick start, this is where the “magic” gets unpacked. Read the pages in order:

  1. Architecture — the methodology. What each phase does, why Dapper combines white-box source analysis with black-box exploitation, and the No Exploit, No Report principle that keeps the final report free of false positives.
  2. The agent pipeline — the orchestration. How dozens of specialist agents are paired into vuln→exploit pipelines, run concurrently, and survive crashes on a Temporal workflow.
  3. MCP & tooling — the hands. The MCP servers and external tools agents use to actually act on the target.
Methodology
Architecture
The four phases: Recon, Analysis, Exploitation, Reporting.
Orchestration
The agent pipeline
Parallel specialist agents on a durable Temporal workflow.
Tooling
MCP & tooling
How agents act: Playwright, TOTP, deliverables, recon tools.

Table of contents