Features
An agent with a shell can invoke Ghidra, angr, a fuzzer, and a debugger. Invocation is not analysis. Dr.Binary is the environment that turns those tools into something an agent can actually investigate with.
01 · Foundation
A binary your agent can browse, read, and grep.
Before the agent asks its first question, Dr.Binary disassembles and decompiles every function, recovers call relationships including indirect calls, gives functions meaningful names and descriptions, and indexes everything for search. The result is a workspace that looks like a codebase.
Example
“Find the authentication logic”
semantic search → inspect candidate functions → follow their callers
02 · Execution
Inspect what a program does. Explore what it could do.
One execution environment for concrete and symbolic reasoning. The agent runs the program, steps through instructions, inspects registers and memory, then marks inputs symbolic, constrains paths, and solves for the input that reaches a target. Interactively, or as a Python script when the analysis repeats.
Example
“Find an input that reaches this security-sensitive code path”
static hypothesis → concrete probe → symbolic inputs → solved input
03 · Workflow
The diff engine finds changes. The agent works out what they mean.
Two binaries go in. A navigable set of modified, added, and removed functions comes out, with function-level matching and decompiled before-and-after code, as machine-readable artifacts the agent can search and drill into rather than a single unstructured report.
Example
“What changed in this update, and could it be security-relevant?”
detect → explain → assess reachability → qualify
Design principles
01
Meaningful analysis operations, not raw commands to reconcile.
02
Program entities, representations, and artifacts connected across tools.
03
Results preserved and reused through an iterative investigation.
04
Conclusions tied to precise, reproducible program facts.
05
Hosted, external, and locally deployed agents all supported.
Underneath
Dr.Binary runs the standard reverse-engineering stack plus two proprietary engines, all in the cloud with nothing to install. What you are paying for is the layer that connects them: one program representation, results computed once and reused, and findings that stay tied to the facts that produced them.
Where it runs
Claude Code, OpenAI Codex, and VS Code plugins, or any MCP client pointed at mcp.drbinary.ai. OAuth sign-in, nothing to install server-side.
Set up MCPAsk in plain language and watch Dr.Binary choose and run the analyses. Build custom agents that encode your own methodology.
Open the workspaceBring your own frontier, enterprise, or local models. Private-cloud deployment, or the AirGap appliance with no internet required.
About AirGapFree to start. Connect over MCP in two commands, or open the browser workspace.