[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-71986":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":23,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":35,"readmeContent":36,"aiSummary":37,"trendingCount":16,"starSnapshotCount":16,"syncStatus":38,"lastSyncTime":39,"discoverSource":40},71986,"ida-pro-mcp","mrexodia\u002Fida-pro-mcp","mrexodia","AI-powered reverse engineering assistant that bridges IDA Pro with language models through MCP.","https:\u002F\u002Fplugins.hex-rays.com\u002Fmrexodia\u002Fida-pro-mcp",null,"Python",9270,1111,48,24,0,116,251,844,348,40.14,"MIT License",false,"main",[26,27,28,29,30,31,32,33,34],"ai","aiagent","binary-analysis","ida-plugin","ida-pro","mcp","mcp-server","modelcontextprotocol","reverse-engineering","2026-06-12 02:02:57","# IDA Pro MCP\n\nSimple [MCP Server](https:\u002F\u002Fmodelcontextprotocol.io\u002Fintroduction) to allow vibe reversing in IDA Pro.\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F6ebeaa92-a9db-43fa-b756-eececce2aca0\n\nThe binaries and prompt for the video are available in the [mcp-reversing-dataset](https:\u002F\u002Fgithub.com\u002Fmrexodia\u002Fmcp-reversing-dataset) repository.\n\n## Prerequisites\n\n- [Python](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F) (**3.11 or higher**)\n  - Use `idapyswitch` to switch to the newest Python version\n- [IDA Pro](https:\u002F\u002Fhex-rays.com\u002Fida-pro) (8.3 or higher, 9 recommended), **IDA Free is not supported**\n- Supported MCP Client (pick one you like)\n  - [Amazon Q Developer CLI](https:\u002F\u002Faws.amazon.com\u002Fq\u002Fdeveloper\u002F)\n  - [Augment Code](https:\u002F\u002Fwww.augmentcode.com\u002F)\n  - [Claude](https:\u002F\u002Fclaude.ai\u002Fdownload)\n  - [Claude Code](https:\u002F\u002Fwww.anthropic.com\u002Fcode)\n  - [Cline](https:\u002F\u002Fcline.bot)\n  - [Codex](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex)\n  - [Copilot CLI](https:\u002F\u002Fdocs.github.com\u002Fen\u002Fcopilot)\n  - [Crush](https:\u002F\u002Fgithub.com\u002Fcharmbracelet\u002Fcrush)\n  - [Cursor](https:\u002F\u002Fcursor.com)\n  - [Gemini CLI](https:\u002F\u002Fgoogle-gemini.github.io\u002Fgemini-cli\u002F)\n  - [Kilo Code](https:\u002F\u002Fwww.kilocode.com\u002F)\n  - [Kiro](https:\u002F\u002Fkiro.dev\u002F)\n  - [LM Studio](https:\u002F\u002Flmstudio.ai\u002F)\n  - [Opencode](https:\u002F\u002Fopencode.ai\u002F)\n  - [Qodo Gen](https:\u002F\u002Fwww.qodo.ai\u002F)\n  - [Qwen Coder](https:\u002F\u002Fqwenlm.github.io\u002Fqwen-code-docs\u002F)\n  - [Roo Code](https:\u002F\u002Froocode.com)\n  - [Trae](https:\u002F\u002Ftrae.ai\u002F)\n  - [VS Code](https:\u002F\u002Fcode.visualstudio.com\u002F)\n  - [VS Code Insiders](https:\u002F\u002Fcode.visualstudio.com\u002Finsiders)\n  - [Warp](https:\u002F\u002Fwww.warp.dev\u002F)\n  - [Windsurf](https:\u002F\u002Fwindsurf.com)\n  - [Zed](https:\u002F\u002Fzed.dev\u002F)\n  - [Other MCP Clients](https:\u002F\u002Fmodelcontextprotocol.io\u002Fclients#example-clients): Run `ida-pro-mcp --config` to get the JSON config for your client.\n\n## Installation (Claude Code)\n\nTo install the headless IDA Pro MCP in Claude Code:\n\n```bash\nclaude plugin marketplace add mrexodia\u002Fclaude-marketplace\nclaude plugin install ida-pro-mcp@mrexodia\n```\n\nTo update to the latest version:\n\n```bash\nclaude plugin update ida-pro-mcp@mrexodia\n```\n\n**Note**: This requires having idalib activated globally and [uv](https:\u002F\u002Fastral.sh\u002Fuv) installed:\n\n```bash\n# windows\nuv run \"C:\\Program Files\\IDA Professional 9.3\\idalib\\python\\py-activate-idalib.py\"\n# macos\nuv run \"\u002FApplications\u002FIDA Professional 9.3.app\u002FContents\u002FMacOS\u002Fidalib\u002Fpython\u002Fpy-activate-idalib.py\"\n```\n\n## Installation (GUI)\n\n**Note**: the MCP plugin is no longer recommended and will eventually be deprecated. Use `idalib-mcp` instead.\n\nIf you want to configure the MCP server manually from the IDA GUI:\n\n```sh\npip uninstall ida-pro-mcp\npip install https:\u002F\u002Fgithub.com\u002Fmrexodia\u002Fida-pro-mcp\u002Farchive\u002Frefs\u002Fheads\u002Fmain.zip\n```\n\nConfigure the MCP servers and install the IDA Plugin:\n\n```\nida-pro-mcp --install\n```\n\n**Important**: Make sure you completely restart IDA and your MCP client for the installation to take effect. Some clients (like Claude) run in the background and need to be quit from the tray icon.\n\n## Prompt Engineering\n\nLLMs are prone to hallucinations and you need to be specific with your prompting. For reverse engineering the conversion between integers and bytes are especially problematic. Below is a minimal example prompt, feel free to start a discussion or open an issue if you have good results with a different prompt:\n\n```md\nYour task is to analyze a crackme in IDA Pro. You can use the MCP tools to retrieve information. In general use the following strategy:\n\n- Inspect the decompilation and add comments with your findings\n- Rename variables to more sensible names\n- Change the variable and argument types if necessary (especially pointer and array types)\n- Change function names to be more descriptive\n- If more details are necessary, disassemble the function and add comments with your findings\n- NEVER convert number bases yourself. Use the `int_convert` MCP tool if needed!\n- Do not attempt brute forcing, derive any solutions purely from the disassembly and simple python scripts\n- Create a report.md with your findings and steps taken at the end\n- When you find a solution, prompt to user for feedback with the password you found\n```\n\nThis prompt was just the first experiment, please share if you found ways to improve the output!\n\nAnother prompt by [@can1357](https:\u002F\u002Fgithub.com\u002Fcan1357):\n\n```md\nYour task is to create a complete and comprehensive reverse engineering analysis. Reference AGENTS.md to understand the project goals and ensure the analysis serves our purposes.\n\nUse the following systematic methodology:\n\n1. **Decompilation Analysis**\n   - Thoroughly inspect the decompiler output\n   - Add detailed comments documenting your findings\n   - Focus on understanding the actual functionality and purpose of each component (do not rely on old, incorrect comments)\n\n2. **Improve Readability in the Database**\n   - Rename variables to sensible, descriptive names\n   - Correct variable and argument types where necessary (especially pointers and array types)\n   - Update function names to be descriptive of their actual purpose\n\n3. **Deep Dive When Needed**\n   - If more details are necessary, examine the disassembly and add comments with findings\n   - Document any low-level behaviors that aren't clear from the decompilation alone\n   - Use sub-agents to perform detailed analysis\n\n4. **Important Constraints**\n   - NEVER convert number bases yourself - use the int_convert MCP tool if needed\n   - Use MCP tools to retrieve information as necessary\n   - Derive all conclusions from actual analysis, not assumptions\n\n5. **Documentation**\n   - Produce comprehensive RE\u002F*.md files with your findings\n   - Document the steps taken and methodology used\n   - When asked by the user, ensure accuracy over previous analysis file\n   - Organize findings in a way that serves the project goals outlined in AGENTS.md or CLAUDE.md\n```\n\nLive stream discussing prompting and showing some real-world malware analysis:\n\n[![](https:\u002F\u002Fimg.youtube.com\u002Fvi\u002FiFxNuk3kxhk\u002F0.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=iFxNuk3kxhk)\n\n## Tips for Enhancing LLM Accuracy\n\nLarge Language Models (LLMs) are powerful tools, but they can sometimes struggle with complex mathematical calculations or exhibit \"hallucinations\" (making up facts). Make sure to tell the LLM to use the `int_convert` MCP tool and you might also need [math-mcp](https:\u002F\u002Fgithub.com\u002FEthanHenrickson\u002Fmath-mcp) for certain operations.\n\nAnother thing to keep in mind is that LLMs will not perform well on obfuscated code. Before trying to use an LLM to solve the problem, take a look around the binary and spend some time (automatically) removing the following things:\n\n- String encryption\n- Import hashing\n- Control flow flattening\n- Code encryption\n- Anti-decompilation tricks\n\nYou should also use a tool like Lumina or FLIRT to try and resolve all the open source library code and the C++ STL, this will further improve the accuracy.\n\n## Transports & Headless MCP\n\nYou can run an SSE server to connect to the user interface like this:\n\n```sh\nuv run ida-pro-mcp --transport http:\u002F\u002F127.0.0.1:8744\u002Fsse\n```\n\nAfter installing [`idalib`](https:\u002F\u002Fdocs.hex-rays.com\u002Fcore\u002Fidalib\u002Fgetting-started) you can also run a headless MCP server. You can start with an initial binary:\n\n```sh\nuv run idalib-mcp --host 127.0.0.1 --port 8745 path\u002Fto\u002Fexecutable\n```\n\nOr start without a binary and open\u002Fclose arbitrary files later with `idalib_open(...)` \u002F `idalib_close(...)`:\n\n```sh\nuv run idalib-mcp --host 127.0.0.1 --port 8745\n```\n\nFor stdio-based clients, use:\n\n```sh\nuv run idalib-mcp --stdio\n```\n\n_Note_: The `idalib` feature was contributed by [Willi Ballenthin](https:\u002F\u002Fgithub.com\u002Fwilliballenthin).\n\n## Headless idalib Session Model\n\n`idalib-mcp` is a supervisor that keeps each open database in its own idalib worker process. Starting without an `input_path` is supported; use `idalib_open(input_path, ...)` to open databases dynamically and `idalib_close(session_id)` to close them. This allows one headless MCP server to work with arbitrary files over its lifetime.\n\nIf the requested IDB is already open in a GUI IDA instance running the plugin, `idalib-mcp` will use that GUI instance instead of spawning a duplicate headless worker. If the GUI instance later disappears, the next routed request reopens the database in a headless worker when possible. Unsaved GUI-only changes must be saved first if they should be visible after fallback.\n\nTools target either the database bound to the current MCP context or an explicit `database` argument.\n\n```sh\nuv run idalib-mcp --stdio --max-workers 4\n```\n\nTypical flow:\n\n```python\nidalib_open(\"\u002Fpath\u002Fto\u002Fbinary_a.exe\", session_id=\"binary_a\")\nidalib_open(\"\u002Fpath\u002Fto\u002Flibrary.dll\", session_id=\"library\")\n\ndecompile(\"main\", database=\"binary_a\")\nxrefs_to(\"ImportantExport\", database=\"library\")\n```\n\n`database` accepts a session ID, filename, or input path. If omitted, tools use the database bound to the active context.\n\nUse `--isolated-contexts` to enable strict per-transport isolation:\n\n```sh\nuv run idalib-mcp --isolated-contexts --host 127.0.0.1 --port 8745 path\u002Fto\u002Fexecutable\n```\n\n### Why use `--isolated-contexts`?\n\nUse it when multiple agents connect to the same `idalib-mcp` server and you want deterministic context isolation:\n\n- Prevent one agent from changing another agent's active database accidentally.\n- Keep each transport context's default database explicit.\n- Still allow intentional collaboration by passing `database=...` or binding multiple agents to the same session ID.\n\nWhen `--isolated-contexts` is enabled:\n\n- Each transport context has its own binding (`Mcp-Session-Id` for `\u002Fmcp`, `session` for `\u002Fsse`, `stdio:default` for stdio).\n- Unbound contexts fail fast for IDB-dependent tools\u002Fresources unless `database` is provided.\n- `idalib_switch(session_id)` and `idalib_open(...)` bind the caller context only.\n\n### Streamable HTTP behavior\n\nWith `--isolated-contexts`, strict Streamable HTTP session semantics are enabled, including `Mcp-Session-Id` validation.\n\n### Context tools\n\n- `idalib_open(input_path, ...)`: Open binary in a worker and bind it to the active context policy.\n- `idalib_switch(session_id)`: Rebind the active context policy to an existing session.\n- `idalib_current()`: Return the session bound to the active context policy.\n- `idalib_unbind()`: Remove the active context binding.\n- `idalib_list()`: Includes `is_active`, `is_current_context`, `bound_contexts`, backend (`worker` or `gui`), and process IDs.\n\nWorker controls:\n\n- `--max-workers N`: maximum simultaneous database workers (`0` = unlimited, default `4`).\n- `IDA_MCP_MAX_WORKERS`: environment default for `--max-workers`.\n\n\n## MCP Resources\n\n**Resources** represent browsable state (read-only data) following MCP's philosophy.\n\n**Core IDB State:**\n- `ida:\u002F\u002Fidb\u002Fmetadata` - IDB file info (path, arch, base, size, hashes)\n- `ida:\u002F\u002Fidb\u002Fsegments` - Memory segments with permissions\n- `ida:\u002F\u002Fidb\u002Fentrypoints` - Entry points (main, TLS callbacks, etc.)\n\n**UI State:**\n- `ida:\u002F\u002Fcursor` - Current cursor position and function\n- `ida:\u002F\u002Fselection` - Current selection range\n\n**Type Information:**\n- `ida:\u002F\u002Ftypes` - All local types\n- `ida:\u002F\u002Fstructs` - All structures\u002Funions\n- `ida:\u002F\u002Fstruct\u002F{name}` - Structure definition with fields\n\n**Lookups:**\n- `ida:\u002F\u002Fimport\u002F{name}` - Import details by name\n- `ida:\u002F\u002Fexport\u002F{name}` - Export details by name\n- `ida:\u002F\u002Fxrefs\u002Ffrom\u002F{addr}` - Cross-references from address\n\n## Core Functions\n\n- `lookup_funcs(queries)`: Get function(s) by address or name (auto-detects, accepts list or comma-separated string).\n- `int_convert(inputs)`: Convert numbers to different formats (decimal, hex, bytes, ASCII, binary).\n- `list_funcs(queries)`: List functions (paginated, filtered).\n- `list_globals(queries)`: List global variables (paginated, filtered).\n- `imports(offset, count)`: List all imported symbols with module names (paginated).\n- `decompile(addr)`: Decompile function at the given address.\n- `disasm(addr)`: Disassemble function with full details (arguments, stack frame, etc).\n- `xrefs_to(addrs)`: Get all cross-references to address(es).\n- `xrefs_to_field(queries)`: Get cross-references to specific struct field(s).\n- `callees(addrs)`: Get functions called by function(s) at address(es).\n\n## Modification Operations\n\n- `set_comments(items)`: Set comments at address(es) in both disassembly and decompiler views.\n- `patch_asm(items)`: Patch assembly instructions at address(es).\n- `declare_type(decls)`: Declare C type(s) in the local type library.\n- `define_func(items)`: Define function(s) at address(es). Optionally specify `end` for explicit bounds.\n- `define_code(items)`: Convert bytes to code instruction(s) at address(es).\n- `undefine(items)`: Undefine item(s) at address(es), converting back to raw bytes. Optionally specify `end` or `size`.\n\n## Memory Reading Operations\n\n- `get_bytes(addrs)`: Read raw bytes at address(es).\n- `get_int(queries)`: Read integer values using ty (i8\u002Fu64\u002Fi16le\u002Fi16be\u002Fetc).\n- `get_string(addrs)`: Read null-terminated string(s).\n- `get_global_value(queries)`: Read global variable value(s) by address or name (auto-detects, compile-time values).\n\n## Stack Frame Operations\n\n- `stack_frame(addrs)`: Get stack frame variables for function(s).\n- `declare_stack(items)`: Create stack variable(s) at specified offset(s).\n- `delete_stack(items)`: Delete stack variable(s) by name.\n\n## Structure Operations\n\n- `read_struct(queries)`: Read structure field values at specific address(es).\n- `search_structs(filter)`: Search structures by name pattern.\n\n## Debugger Operations (Extension)\n\nDebugger tools are hidden by default. Enable with `?ext=dbg` query parameter:\n\n```\nhttp:\u002F\u002F127.0.0.1:13337\u002Fmcp?ext=dbg\n```\n\n**Control:**\n- `dbg_start()`: Start debugger process.\n- `dbg_exit()`: Exit debugger process.\n- `dbg_continue()`: Continue execution.\n- `dbg_run_to(addr)`: Run to address.\n- `dbg_step_into()`: Step into instruction.\n- `dbg_step_over()`: Step over instruction.\n\n**Breakpoints:**\n- `dbg_bps()`: List all breakpoints.\n- `dbg_add_bp(addrs)`: Add breakpoint(s).\n- `dbg_delete_bp(addrs)`: Delete breakpoint(s).\n- `dbg_toggle_bp(items)`: Enable\u002Fdisable breakpoint(s).\n\n**Registers:**\n- `dbg_regs()`: All registers, current thread.\n- `dbg_regs_all()`: All registers, all threads.\n- `dbg_regs_remote(tids)`: All registers, specific thread(s).\n- `dbg_gpregs()`: GP registers, current thread.\n- `dbg_gpregs_remote(tids)`: GP registers, specific thread(s).\n- `dbg_regs_named(names)`: Named registers, current thread.\n- `dbg_regs_named_remote(tid, names)`: Named registers, specific thread.\n\n**Stack & Memory:**\n- `dbg_stacktrace()`: Call stack with module\u002Fsymbol info.\n- `dbg_read(regions)`: Read memory from debugged process.\n- `dbg_write(regions)`: Write memory to debugged process.\n\n## Advanced Analysis Operations\n\n- `py_eval(code)`: Execute arbitrary Python code in IDA context (returns dict with result\u002Fstdout\u002Fstderr, supports Jupyter-style evaluation).\n- `analyze_funcs(addrs)`: Comprehensive function analysis (decompilation, assembly, xrefs, callees, callers, strings, constants, basic blocks).\n\n## Pattern Matching & Search\n\n- `find_regex(queries)`: Search strings with case-insensitive regex (paginated).\n- `find_bytes(patterns, limit=1000, offset=0)`: Find byte pattern(s) in binary (e.g., \"48 8B ?? ??\"). Max limit: 10000.\n- `find_insns(sequences, limit=1000, offset=0)`: Find instruction sequence(s) in code. Max limit: 10000.\n- `find(type, targets, limit=1000, offset=0)`: Advanced search (immediate values, strings, data\u002Fcode references). Max limit: 10000.\n\n## Control Flow Analysis\n\n- `basic_blocks(addrs)`: Get basic blocks with successors and predecessors.\n\n## Type Operations\n\n- `set_type(edits)`: Apply type(s) to functions, globals, locals, or stack variables.\n- `infer_types(addrs)`: Infer types at address(es) using Hex-Rays or heuristics.\n\n## Export Operations\n\n- `export_funcs(addrs, format)`: Export function(s) in specified format (json, c_header, or prototypes).\n\n## Graph Operations\n\n- `callgraph(roots, max_depth)`: Build call graph from root function(s) with configurable depth.\n\n## Batch Operations\n\n- `rename(batch)`: Unified batch rename operation for functions, globals, locals, and stack variables (accepts dict with optional `func`, `data`, `local`, `stack` keys).\n- `patch(patches)`: Patch multiple byte sequences at once.\n- `put_int(items)`: Write integer values using ty (i8\u002Fu64\u002Fi16le\u002Fi16be\u002Fetc).\n\n**Key Features:**\n\n- **Type-safe API**: All functions use strongly-typed parameters with TypedDict schemas for better IDE support and LLM structured outputs\n- **Batch-first design**: Most operations accept both single items and lists\n- **Consistent error handling**: All batch operations return `[{..., error: null|string}, ...]`\n- **Cursor-based pagination**: Search functions return `cursor: {next: offset}` or `{done: true}` (default limit: 1000, enforced max: 10000 to prevent token overflow)\n- **Performance**: Strings are cached with MD5-based invalidation to avoid repeated `build_strlist` calls in large projects\n\n## Development\n\nAdding new features is a super easy and streamlined process. All you have to do is add a new `@tool` function to the modular API files in `src\u002Fida_pro_mcp\u002Fida_mcp\u002Fapi_*.py` and your function will be available in the MCP server without any additional boilerplate! Below is a video where I add the `get_metadata` function in less than 2 minutes (including testing):\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F951de823-88ea-4235-adcb-9257e316ae64\n\nTo test the MCP server itself:\n\n```sh\nnpx -y @modelcontextprotocol\u002Finspector\n```\n\nThis will open a web interface at http:\u002F\u002Flocalhost:5173 and allow you to interact with the MCP tools for testing.\n\nFor testing I create a symbolic link to the IDA plugin and then POST a JSON-RPC request directly to `http:\u002F\u002Flocalhost:13337\u002Fmcp`. After [enabling symbolic links](https:\u002F\u002Flearn.microsoft.com\u002Fen-us\u002Fwindows\u002Fapps\u002Fget-started\u002Fenable-your-device-for-development) you can run the following command:\n\n```sh\nuv run ida-pro-mcp --install\n```\n\nGenerate the changelog of direct commits to `main`:\n\n```sh\ngit log --first-parent --no-merges 1.2.0..main \"--pretty=- %s\"\n```\n","mrexodia\u002Fida-pro-mcp 是一个基于AI的逆向工程辅助工具，通过MCP协议将IDA Pro与多种语言模型连接起来。该项目的核心功能是提供了一个简单的MCP服务器，使得用户能够在IDA Pro中利用AI进行二进制分析和逆向工程。它支持多种MCP客户端，包括Claude Code、Qwen Coder等，极大提高了逆向工程的效率和准确性。适用于需要对复杂软件进行深入分析的安全研究人员、开发人员以及逆向工程师使用。项目采用Python编写，并且要求Python 3.11及以上版本和IDA Pro 8.3及以上版本（推荐9.0）。",2,"2026-06-11 03:39:50","high_star"]