How fuzzinglabs/fuzzforge_ai works
Project Type and Purpose
• FuzzForge OSS is an open‐source, AI‐driven security research orchestration runtime.
• It implements a microservice‐style architecture in Python, coordinating containerized security modules via the Model Context Protocol (MCP).
File Structure
• Top‐level “services” and libraries:
• fuzzforge-cli – CLI/front-end
• fuzzforge-mcp – MCP server (API layer)
• fuzzforge-runner – Execution engine (business logic)
• fuzzforge-common – Shared abstractions (container engines, storage, queues, workflows)
• fuzzforge-types – Typed schemas for resources, modules, workflows
• fuzzforge-modules & fuzzforge-modules-sdk – Containerized tooling modules and SDK to build them
• Each security module lives under fuzzforge-modules as a Docker/Podman image.
README for Architecture Insights
AI Agent → MCP Server → Runner → Container Engine → Modules
• Highlights MCP RPC methods (list_modules, execute_module, start_continuous_module).
Main Components & Their Interactions
a. AI Agent (Copilot, Claude, etc.)
• Speaks MCP over STDIO or plugin interface
b. FuzzForge CLI
• Front‐end for human operators and MCP plugin installer
• Forwards commands to the MCP server
c. FuzzForge MCP Server
• Exposes MCP RPC endpoints
• Uses fuzzforge-common and fuzzforge-types
• Delegates execution requests to the Runner
d. FuzzForge Runner
• Core orchestrator & executor
• Uses container‐engine wrappers (Docker/Podman) from fuzzforge-common
• Manages sessions, long‐running continuous fuzzers, result collection
• Stores artifacts via fuzzforge-common’s storage (local FS or S3)
e. Container Engine (Docker / Podman)
• Actually runs each module in isolation
f. Modules (Module A, B, C, …)
• Self‐contained security tools (static analyzers, fuzzers, crash analyzers)
• Built with the Python SDK in fuzzforge-modules-sdk
g. Storage & Queues
• Optional temporal queues for asynchronous job dispatch
Architectural Patterns & Design Principles
• Microservice separation: CLI, API server (MCP), executor (Runner), modules as isolated containers
• Plugin/SDK extensibility: fuzzforge-modules-sdk for building new modules
• Typed contracts: MCP payloads described in fuzzforge-types
• Container orchestration abstraction: unified Docker/Podman engines in fuzzforge-common
Detail Without Overthinking
• How fuzzforge-common and fuzzforge-types underpin both the MCP server and the Runner.
• Don’t attempt to show every subpackage—focus on the high-level services, shared libraries, execution engine, container infra, and modules.
• Reflects the actual directory‐based microservice decomposition
• Mirrors the README’s architecture section
• Clearly communicates component responsibilities, data flows, and extensibility points
• Serves as a solid blueprint for onboarding new engineers or designing enhancements