LeechCore MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@LeechCore MCPRead 16 bytes from physical address 0x1000 as a hexdump."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
LeechCore MCP
Read-only physical-memory access over MCP. The native backend calls LeechCore's LcCreate, LcRead, and LcClose directly. A raw-file backend supports offline inspection without a native runtime.
The server exposes nine tools over stdio. Targets, address ranges, and read limits are set in a local JSON file; MCP clients cannot change them. There is no process lookup, virtual-address translation, memory-write API, or HTTP listener.
Requirements
Windows x64 and 64-bit Python 3.10 or later, including
pythonw.exe.PowerShell 5.1 or later for the setup script.
For native access: a trusted x64
leechcore.dlland its matching device dependencies. DLLs, drivers, and firmware are not included.
Related MCP server: memscope-mcp
Install
Clone the repository and run setup:
git clone https://github.com/0xTracepacket/leechcore-mcp.git
cd leechcore-mcp
.\scripts\setup.ps1To select a Python installation:
.\scripts\setup.ps1 -Python 'C:\Python311\python.exe'Setup creates .venv, installs hash-locked dependencies, generates a 64 KiB synthetic demo, and prints MCP client configuration. It preserves an existing demo configuration and does not edit client settings or open a device.
Connect a client
Copy the TOML or JSON snippet printed by setup into your MCP client's settings. To generate it again:
.\.venv\Scripts\python.exe -m leechcore_mcp client-config --config .\demo\config.json --format codex
.\.venv\Scripts\python.exe -m leechcore_mcp client-config --config .\demo\config.json --format jsonRestart the client connection, then call these tools in order:
Tool | Arguments | Result |
|
| Backend, allowed ranges, limits, and session state. |
|
| Opens the configured demo file. |
|
|
|
|
| Releases the session. |
The server waits for MCP messages on stdin when started manually. Use doctor for command-line diagnostics, not serve.
Native access
Copy examples/leechcore.example.json to a private configuration file. Set dll_path, select device: "fpga" or a local dump-file path, and replace allowed_ranges with the physical ranges you are authorized to inspect. Relative paths resolve from the configuration directory.
Native reads acquire complete 4096-byte pages and return the requested slice. The entire enclosing page range must be allowed: one byte at 0x1001 requires 0x1000 through 0x1fff. Budgets count requested bytes, not native bus traffic.
.\.venv\Scripts\python.exe -m leechcore_mcp --config .\target.json doctor
.\.venv\Scripts\python.exe -m leechcore_mcp client-config --config .\target.json --format codexdoctor checks files, configuration, and DLL architecture without loading the DLL. memory_open loads it in the acquisition worker. Use a trusted runtime: the read-only tool API does not restrict code inside an external DLL or driver.
See setup for configuration fields, client integration, and the optional native check.
Tools
All addresses are 0x-prefixed strings. File addresses are byte offsets; native addresses are physical addresses.
Tool | Parameters |
| None. |
| None. |
| None. |
|
|
|
|
|
|
|
|
|
|
|
|
The leechcore://policy resource exposes the configured ranges and limits without opening the source. Architecture covers types, bounds, timeouts, and output semantics; demo calls provide sample arguments.
Test and build
.\.venv\Scripts\python.exe -m unittest discover -s tests -v
.\.venv\Scripts\python.exe -m pip check
.\.venv\Scripts\python.exe .\scripts\build_bundle.pyThe builder writes a source ZIP and SHA-256 checksum to dist, with a per-file manifest inside the archive. Runtime binaries, memory files, logs, private configurations, and virtual environments are excluded.
For development, install .[dev], then run ruff check . and ruff format --check .. CI runs lint, formatting, and tests on Windows with Python 3.10 and 3.14.
Release verification passed 97 tests on Python 3.10.5 and 3.14.0, plus LeechCore 2.23.3 native reads and an MCP workflow against synthetic data. FPGA and live-target access have not been tested. See the verification record.
Data handling
The MCP client can forward returned memory to its provider and retain it in transcripts. Check that data path before connecting real memory. Treat returned bytes and decoded strings as data, not instructions. Optional JSONL audit logs contain operation metadata, not raw memory.
Reference
This server cannot be deployed
Maintenance
Related MCP Connectors
Read-only MCP access to authorized Vocci sessions, notes, files, and memory search.
Reasoning, code, anti-deception, memory harness MCP tools. Stdio or HTTPS api.ejentum.com/mcp
A paid remote MCP for agent memory MCP, built to return verdicts, receipts, usage logs, and audit-re
- KogniteOAuthdev.kognite
Hosted agent memory: store, search, and recall facts across sessions from any MCP client.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceProvides safe, read-only access to memory analysis and debugging functionality through the Model Context Protocol, allowing users to examine computer memory for software development, security research, and educational purposes.63MIT
- AlicenseNot gradedqualityAmaintenanceEnables AI agents to perform low-level Windows process memory research, including process attachment, memory scanning, reading/writing, pointer chasing, remote code execution, and inline hooking via MCP tools and Lua scripting.2MIT
- AlicenseNot gradedqualityCmaintenanceAn MCP server for Windows process memory inspection/editing and crash dump analysis, enabling live memory scanning, patching, pointer resolution, disassembly, and offline dump examination via structured tool calls.MIT
- FlicenseNot gradedqualityCmaintenanceA Model Context Protocol (MCP) server for memory forensics analysis using MemProcFS.-