Linux内核邮件列表解析服务

LKML MCP Server

Read the Linux kernel mailing lists from inside your AI assistant — no browser tabs, no git format-patch gymnastics, no scrolling through lore.kernel.org's HTML.

Ask "what did the reviewers say about my RISC-V series?" or "grab v3 of this patchset so I can git am it" and get a structured answer back. The server talks to lore.kernel.org (and any compatible public-inbox archive) and hands your assistant clean, parsed threads instead of raw HTML.

Why

Following kernel review happens over email, and email tooling is hostile to LLMs: deeply nested quotes, MIME soup, bot noise, threads scattered across dozens of replies. This server does the unglamorous parsing so your assistant can reason about the actual conversation — who reviewed what, which tags landed, what changed between versions.

Works against any list on lore.kernel.org without hardcoding names — lkml, netdev, linux-riscv, devicetree, kvm, and the rest — plus per-inbox instances like inbox.sourceware.org for GCC/Glibc/GDB.

Tools

In practice

You don't call these tools directly; you talk to your assistant and it picks the right one. A few things you might say:

• "Pull up the thread for 20251111105634.1684751-1-lzampier@redhat.com and tell me if anyone objected."
• "Summarize the review tags on this series — did it get enough Reviewed-bys to merge?"
• "Find patches touching the riscv subsystem from the last week."
• "Compare v2 and v3 of this patchset and show me which files changed."
• "Download the whole series as an mbox so I can apply it locally."

Installation

Claude Code, one line:

claude mcp add -s user lkml uvx -- --from "git+https://github.com/zampierilucas/lkml-mcp" lkml-mcp

Any other MCP client takes the same uvx command — drop this into its config:

{
  "mcpServers": {
    "lkml": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/zampierilucas/lkml-mcp", "lkml-mcp"]
    }
  }
}

git clone https://github.com/zampierilucas/lkml-mcp.git
cd lkml-mcp && pip install -e .

Tool reference

How it works

The server fetches from lore.kernel.org using stable message-ID URLs with automatic list detection:

• Threads: https://lore.kernel.org/r/{message-id}/t.mbox.gz (compressed mbox)
• Raw messages: https://lore.kernel.org/r/{message-id}/raw (RFC822)

The /r/ (redirect) endpoint forwards to /all/, which gives cross-list access without hardcoding mailing-list names — so the same URL works whether the message lives on lkml, linux-riscv, netdev, or anywhere else. Messages are parsed into the structured fields above for easy consumption by LLM tools.

Configuration

Different archive (LKML_BASE_URL)

Defaults to https://lore.kernel.org. Point it elsewhere by setting LKML_BASE_URL:

{
  "mcpServers": {
    "lkml": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/zampierilucas/lkml-mcp", "lkml-mcp"],
      "env": { "LKML_BASE_URL": "https://custom-lore-instance.org" }
    }
  }
}

Local:

export LKML_BASE_URL="https://custom-lore-instance.org"
python -m lkml_mcp.server

Transport (stdio / SSE / Streamable HTTP)

Runs on stdio by default; can also be exposed over HTTP so multiple clients share one process. Streamable HTTP is stateless under the 2026-07-28 spec — requests can hit any instance behind a load balancer. SSE is deprecated and kept only for older clients.

Run as a daemon:

lkml-mcp --transport streamable-http --host 0.0.0.0 --port 8772
# or
lkml-mcp --transport sse --host 0.0.0.0 --port 8772

Client endpoints:

• Streamable HTTP: http://<host>:8772/mcp
• SSE: http://<host>:8772/sse (POSTs to /messages/)

Attach the running daemon to Claude Code:

claude mcp add -s user -t http lkml http://127.0.0.1:8772/mcp
# or for SSE:
claude mcp add -s user -t sse lkml http://127.0.0.1:8772/sse

Multiple archives at once

Run more than one instance to read different archives side by side — say, the kernel lists on lore.kernel.org and GCC/Glibc on inbox.sourceware.org:

{
  "mcpServers": {
    "lkml": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/zampierilucas/lkml-mcp", "lkml-mcp"]
    },
    "sourceware": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/zampierilucas/lkml-mcp", "lkml-mcp"],
      "env": { "LKML_BASE_URL": "https://inbox.sourceware.org" }
    }
  }
}

• lkml → kernel lists (no inbox parameter needed)
• sourceware → GCC, Glibc, GDB, binutils (inbox parameter required)

Instance types

The server auto-detects which kind of public-inbox instance it's talking to:

Common sourceware inboxes: gcc, gcc-patches, libc-alpha, gdb-patches, binutils.