Should you use this IDA bridge?
IDA Pro MCP's own README now recommends the Official Hex-Rays IDA MCP Server instead. That changes the starting point for a new installation: investigate the maintainer's recommendation before committing to this integration. The supplied sources do not establish the official server's features, compatibility, or migration steps, so this review does not rank the two.
For an existing IDA workflow, the relevant question is whether this bridge's documented tools fit your task and whether they work in your environment. The README describes decompilation, disassembly, cross-references, types, searches, and call graphs, alongside tools that rename symbols, add comments, patch bytes, and save databases. These are documented capabilities, not workloads we verified in the recorded sandbox run.
What our recorded run establishes
Our recorded run took place on September 28, 2026, at commit c133c3853faa111a9b00ee615c013b720d0c4acd, whose recorded commit date is September 26, 2026. No new execution was performed for this correction.
The configuration records 3 CPUs, 8 GB of memory, and the image ghcr.io/astral-sh/uv:python3.12-bookworm. The install output identifies Python 3.11.2 at /work/home/venv; the image name alone does not establish the runtime version.
The install step exited 0 in 19.670 seconds, rounded to 20 seconds. The record lists 37 dependencies and 54 MB. The build step exited 0 in 4.138 seconds, rounded to 4 seconds, with no retained build output. The exact commands and an installed-binary check are absent from the supplied record. Successful steps therefore do not prove that the CLI was installed, launched, or connected to IDA.
The test command exited 1 after 7.439 seconds, rounded to 7 seconds. The retained pytest output reports 141 passed and 30 errors in 2.51 seconds. The parser records 0 failed, 0 skipped, and 171 total. Treat these as recorded summaries, not an independently reconciled count of unique tests or a coverage measure.
Several collection tracebacks explicitly show ModuleNotFoundError for jsonschema. That identifies a missing dependency for those errors. The tail does not show the causes of all 30 errors, and it cannot establish whether the operational bridge works or fails. A useful next step is to resolve the shown test dependency and rerun the intended suite in a configured IDA environment.
The audit record contains a pip-audit advisory total of 0, with unknown coverage. It does not support a claim that every dependency was inspected or that the software is safe. The checkout metrics were 213 files, approximately 35,655 source lines, and 6 MB, with one CI workflow, a tests directory, and no detected Dockerfile or Compose file.
Setup extends beyond Python dependencies
The README requires Python 3.11 or higher and IDA Pro 8.3 or higher, recommends IDA Pro 9, and excludes IDA Free. Its idalib configuration requires global activation and uv. It supplies client setup paths for Claude Code, Codex, and Kimi Code, plus generated configuration through ida-pro-mcp --config for other clients.
The older GUI MCP plugin is no longer recommended in the README and is intended for eventual deprecation in favor of idalib-mcp. GUI installation requires completely restarting IDA and the MCP client, including clients that remain in the system tray. These requirements sit outside the recorded dependency footprint.
Named sessions need lifecycle management
The documented headless supervisor keeps each database in a separate worker process. Calls require an explicit database session ID returned by opening or listing sessions; a filename or path is not accepted as that argument.
Workers can outlive their supervisor and be adopted by a later supervisor. The documented defaults are four workers and a one-hour idle timeout. idb_close can release an owned worker early, while adopted GUI or worker instances are detached rather than killed. Plan for database locks and persistent processes when ending a client session. These lifecycle details come from the README, not a measured concurrency test.
Review changes and model conclusions
The documented write tools can change names and types, patch bytes, and alter stack variables. py_eval executes arbitrary Python inside IDA. Debugger tools are hidden by default and enabled through the dbg extension; they include execution control and memory writes.
For an initial evaluation, use a copy of the database, keep HTTP listeners on localhost, and review changes before retaining them. This is an operating recommendation based on the tools' documented side effects, not a security finding from our run.
The README also warns that models struggle with obfuscated code. It recommends addressing string encryption, import hashing, control-flow flattening, code encryption, and anti-decompilation tricks before relying on model analysis. Its prompting guidance calls for checking decompilation against assembly when necessary and using int_convert for number conversions. No measured accuracy or time-saving result is available here.
Read issue reports with their revision and status
Issue 506 reported call-graph edges that incorrectly treated jumps as calls at revision 0b5f7ae. The supplied issue page records it as closed as completed through pull request 507 on September 20, 2026. It should not be presented as an unresolved defect. That closure alone does not verify the behavior of a particular installed revision.
Issue 510 reports large integer serialization failures in BigInt-aware clients at revision 0b5f7ae. Issue 497 reports GUI discovery failures on Windows 11 with IDA Pro 9.3. The supplied pages show both reports open. They are user-reported cases, not failures reproduced in our sandbox, and they do not establish that every client or installation is affected.
The supplied release page identifies version 1.4.0 and lists contributions including an idalib server and debugger functions. Release notes and issue activity show development participation; they do not measure adoption, support responsiveness, or compatibility with your environment.
How to read the scores
These are editorial ratings of the supplied evidence, not benchmark results. The scale is 1 for substantial barriers or very limited support, 2 for significant prerequisites or evidence gaps, 3 for useful documented support with material limits, 4 for detailed support across the assessed needs, and 5 for broad support corroborated by relevant operational evidence. Unknown adoption and compatibility do not earn higher scores.
Setup scores 2 because licensed IDA, activation, client configuration, and restarts extend beyond dependency installation. Documentation scores 4 because the README details setup, session lifecycle, tools, and model limitations, although we did not validate those procedures end to end. Community scores 3 because the supplied release and issue pages show contributors and maintainer participation without establishing support reliability. Maturity scores 3 because the documented interface and release history support evaluation, while operational compatibility remains unverified by this run.
For an existing installation, pin the revision, check the installed command, connect to licensed IDA, and try one inspection task on a disposable database before allowing modifications. For a new installation, begin with the maintainer's recommendation and establish whether that server meets the same needs.