Ghidra · plugin
GhidrAssist setup with a custom base URL
GhidrAssist puts an LLM inside Ghidra to explain a function, chat and apply bulk renames. A base URL and a model id run it on Kimi K3.1 Pro.
What it does
- Explain a function, or chat with that function as context.
- Agent mode calls tools and writes findings.
- Bulk renames you apply yourself.
- A graph of per-function summaries.
Install
- If needed, copy the release ZIP to
Ghidra_Install/Extensions/Ghidra. - In Ghidra,
FileInstall Extension, and enable GhidrAssist. - Open a binary in the CodeBrowser, then
FileConfigureMiscellaneousand enable GhidrAssist. - Open it from
WindowGhidraAssistPlugin.
Add the model
In GhidrAssist's settings, under LLM Providers, click Add and fill in:
| Field | Value |
|---|---|
| Name | Silk Compute |
| Type | OpenAI Platform API |
| Model | kimi-k3.1-pro-uncensored |
| URL | https://api.silkcompute.com/v1 |
| API key | Your key |
Set it as the active provider. Use deepseek-v4.1-flash-1017-uncensored when the graph summarizes every function.
jtang613/GhidrAssist README, read November 18, 2026.
Use
Agent mode needs tool calling. All five models in the comparison have it.
GhidrAssistMCP
GhidrAssistMCP serves Ghidra over MCP. It needs Ghidra 12.0 or 12.1 on Java 25, and listens on port 8080.
endpointstext
Streamable HTTP: http://127.0.0.1:8080/mcp (preferred)
SSE: http://127.0.0.1:8080/sse
SSE messages: http://127.0.0.1:8080/messageterminalshell
claude mcp add --transport http ghidrassist http://127.0.0.1:8080/mcpIt also runs inside analyzeHeadless. Model settings are on the MCP setup page.
jtang613/GhidrAssistMCP README, read November 18, 2026.