REA in ChatGPT and Claude, without the local setup.
One authenticated MCP endpoint with the analysis engines already installed. Upload software you are authorized to examine and ask your assistant about it. Request yours now; we get in touch and send your connection details.
Illustrative flow. Your tools and scope follow your audit.
The overview
One endpoint. The engines already installed.
REA runs as a local stdio MCP server, so hosted clients such as ChatGPT web and Claude’s remote connectors cannot reach it, and its native analysis engines are yours to install. We are preparing an independent hosted endpoint that removes both steps: a remote, authenticated REA MCP server with Ghidra, JADX and Kuna ready, restricted to static analysis of files you upload. Endpoints are prepared per request and nothing is charged in this early-access phase.
Why a hosted endpoint
The REA project ships its MCP server over local stdio: the client starts a process on the machine that holds the target. That works in Claude Code, Cursor and Codex on your laptop. ChatGPT web, Claude’s custom connectors and other hosted assistants connect only to a remote HTTP server, so a GitHub URL or an npx command does not get you there.
The second step is the engines. REA wraps bring-your-own tools: Ghidra and a matching JDK for native code, JADX and apktool for Android, Kuna as a static decompiler. Installing and validating them is where most first attempts stop. On a hosted endpoint they are already there and already checked by rea doctor.
What the first release includes
One remote MCP endpoint. Add it as a connector in ChatGPT or Claude, or as a remote server in any MCP client that supports Streamable HTTP. Authentication is per user.
Engines installed. Ghidra 12 with JDK 21, JADX and apktool, and Kuna, all open-source and Apache-2.0 licensed. Hopper and IDA are not included because their licenses do not allow hosted use.
Static analysis only. Tools that run, attach to or instrument a program are not exposed. Your upload is parsed, never executed.
Evidence you can check. REA’s answers keep observed facts, derivations and unknowns distinguishable. The hosted version changes where the tools run, not what they claim.
Isolated sessions. Each user session runs in its own sandboxed container with no outbound network. Uploads are deleted at the end of the retention window or on request.
Upstream first. We contribute fixes to the original project and run the same code. The service adds transport, authentication, isolation and managed engines, nothing else.
How access will work
01
Request below. Tell us which client you use and what you need to analyze. That decides which engines we validate for your endpoint.
02
Hear from us. We contact you to confirm your setup, then send a second email with the endpoint URL, your credentials and the connection steps for your client.
03
Connect and upload. Add the connector, sign in, and upload an application, bundle or binary you are authorized to examine.
04
Ask a bounded question. For example: “Map the export path in this Electron build and list the modules involved. Do not execute anything.” Then review the evidence before acting on it.
Get your dedicated REA MCP
No payment and no account in this phase. We use your answers to prepare your endpoint and your email to get in touch and send your connection details. Reply to any email to be removed.
Boundaries and status
This is an independent service built on REA’s MIT-licensed code by API to Agents, a brand of Neoval GmbH. It is not operated by or affiliated with the REA maintainers. The original project is morluto/rea; our fork is apitoagents/rea.
Use it only on software you own or are permitted to analyze, within the terms that apply to it. The service is not for bypassing access controls, licensing or payment restrictions, or for copying protected code and assets without permission. We will ask you to confirm authorization at upload.
Status on October 17, 2026: requests open, endpoints prepared on request. The existing REA project page describes the local setup that works today.
A few more details
Frequently asked questions
Is the hosted REA endpoint available now?
Requests are open now. We contact each requester personally, then send a second email with the endpoint URL, credentials and connection steps once the dedicated endpoint is ready. Until then, REA works as a local MCP server as described on our REA project page.
Is this an official REA service?
No. REA is an open-source project by morluto and its contributors, released under the MIT license. API to Agents operates this hosted endpoint independently, contributes fixes upstream, and credits the original authors. The maintainers are not responsible for this service.
What does it cost?
Nothing in this early-access phase. No card or account is requested. Pricing for the general service will be announced later and will be per seat or per workspace, not per tool call.
Which analysis engines are included?
Ghidra with JDK 21 for native binaries, JADX and apktool for Android, and Kuna as a static decompiler. All are open-source and licensed for hosted use. Hopper and IDA are excluded because their licenses do not permit it.
Does the service execute the software I upload?
No. Only static analysis tools are exposed. REA's process-capture, Frida, browser and debugger tools are not available on the hosted endpoint. Uploads are parsed inside an isolated container without outbound network access.
Who can see my uploads?
Only your session. Each user runs in a separate container, uploads are stored for a limited retention window, and you can request deletion at any time. Results returned to your assistant are still processed by that assistant's provider under its own terms.