[GH-ISSUE #2927] [Feature] CLI batch mode: run several tool calls in one provider session #1030

New issue

Open

opened 2026-10-15 18:30:29 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-15 18:30:29 +00:00

Owner

Originally created by @matbrgz on GitHub (Oct 13, 2026).
Original GitHub issue: https://github.com/morluto/rea/issues/2927

Before submitting

Searched (batch CLI, CLI session reuse, persistent session, re-import). Related: #1048 (pre-imported Ghidra project), #479 (Hopper window reuse). This asks for a CLI-level batch entry point, which is independent of the provider.
Contribution scope understood. No private data.

Area

CLI

Current limitation

Each CLI invocation is a separate process (docs/cli.md), so with a deep provider every command pays the full import and auto-analysis cost. On a 7.1 MB Delphi PE32 with Ghidra 12.1.4 that is about 70 s per command. An investigation step typically needs 5–20 queries (search, trace, callers, decompile, bytes). Today the only ways to amortize the import are an MCP-capable agent or a hand-written MCP stdio client. We wrote one (tools/rea_session.mjs, ~50 lines using @modelcontextprotocol/client) to run REA from scripts and CI.

Proposed outcome

rea batch <calls.json|-> [--provider ghidra] [--out DIR|--jsonl]:

input: an ordered list of {id, tool, arguments} using the MCP tool names and schemas;
one session: open the target once (first call or --target), run the calls in order, close (optionally saving --snapshot);
output: one Evidence/result per call (JSONL or DIR/<id>.json), a per-call ok|error with the original error contract, and timing;
--stop-on-error / --continue semantics; cancellation closes the provider cleanly.

Use cases

Reproducible reverse-engineering notebooks and CI checks (e.g. "after a vendor update, re-run these 15 queries and diff the Evidence"), without registering REA as an agent MCP server.

Repro of the cost

AUTORIZADOR_VER6630.exe (sha256 0c79af43ce52a9317591e081ebfc9f29f70d5575033b2eb9e177d0828a634c44) → innoextract -e → app/AUTORIZADOR.exe (sha256 1398d55a71a9e4ac04f7a12f5f1bec9d0b4343fbd1c0f8635938d261f18fa363). Running rea decompile <exe> 0x65caac --provider ghidra twice with different addresses takes ~70 s each. The same two queries in one MCP session: ~70 s + ~1 s.

Suggested approach

Reuse the MCP server's tool registry in-process (the CLI already shares workflows with MCP) and drive it from the batch file. No new provider work is needed.

Originally created by @matbrgz on GitHub (Oct 13, 2026). Original GitHub issue: https://github.com/morluto/rea/issues/2927 [![Type: Feature](https://img.shields.io/badge/type-feature-blue.svg?style=flat-square)](https://github.com/morluto/rea) [![Domain: Reverse Engineering](https://img.shields.io/badge/domain-reverse_engineering-blueviolet.svg?style=flat-square)](https://github.com/morluto/rea) [![Status: Proposed](https://img.shields.io/badge/status-proposed-blue.svg?style=flat-square)](https://github.com/morluto/rea) [![Ecosystem: REA](https://img.shields.io/badge/ecosystem-morluto%2Frea-8A2BE2.svg?style=flat-square)](https://github.com/morluto/rea) ### Before submitting - [x] Searched (`batch CLI`, `CLI session reuse`, `persistent session`, `re-import`). Related: #1048 (pre-imported Ghidra project), #479 (Hopper window reuse). This asks for a CLI-level batch entry point, which is independent of the provider. - [x] Contribution scope understood. No private data. ### Area CLI ### Current limitation Each CLI invocation is a separate process (docs/cli.md), so with a deep provider every command pays the full import and auto-analysis cost. On a 7.1 MB Delphi PE32 with Ghidra 12.1.4 that is about **70 s per command**. An investigation step typically needs 5–20 queries (search, trace, callers, decompile, bytes). Today the only ways to amortize the import are an MCP-capable agent or a hand-written MCP stdio client. We wrote one (`tools/rea_session.mjs`, ~50 lines using `@modelcontextprotocol/client`) to run REA from scripts and CI. ### Proposed outcome `rea batch <calls.json|-> [--provider ghidra] [--out DIR|--jsonl]`: - input: an ordered list of `{id, tool, arguments}` using the MCP tool names and schemas; - one session: open the target once (first call or `--target`), run the calls in order, close (optionally saving `--snapshot`); - output: one Evidence/result per call (JSONL or `DIR/<id>.json`), a per-call `ok|error` with the original error contract, and timing; - `--stop-on-error` / `--continue` semantics; cancellation closes the provider cleanly. ### Use cases Reproducible reverse-engineering notebooks and CI checks (e.g. "after a vendor update, re-run these 15 queries and diff the Evidence"), without registering REA as an agent MCP server. ### Repro of the cost `AUTORIZADOR_VER6630.exe` (sha256 `0c79af43ce52a9317591e081ebfc9f29f70d5575033b2eb9e177d0828a634c44`) → `innoextract -e` → `app/AUTORIZADOR.exe` (sha256 `1398d55a71a9e4ac04f7a12f5f1bec9d0b4343fbd1c0f8635938d261f18fa363`). Running `rea decompile <exe> 0x65caac --provider ghidra` twice with different addresses takes ~70 s each. The same two queries in one MCP session: ~70 s + ~1 s. ### Suggested approach Reuse the MCP server's tool registry in-process (the CLI already shares workflows with MCP) and drive it from the batch file. No new provider work is needed.

bsakaguchi added the

area: cli

enhancement

labels

2026-10-15 18:30:29 +00:00

bsakaguchi referenced this issue

2026-10-17 21:07:58 +00:00

[PR #1030] [MERGED] feat(ghidra): support native x86 PE applications on Windows #2594

bsakaguchi added

enhancement

and removed

enhancement