patchwork-mcp-server
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., "@patchwork-mcp-server Show me failing CI checks on patch 14553325"
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.
patchwork-mcp-server
A Model Context Protocol (MCP) server that exposes the Patchwork REST API to LLM-based coding agents (Claude Code, Claude Desktop, Cursor, etc.).
It is a thin, read-only wrapper around /api/1.2/ so you can ask the
agent things like:
"What's the state of my latest
seg6:series on netdev?""Show me failing CI checks on patch 14553325."
"List the cover-letter thread for series 1089407."
"Find the patch with Message-Id
20260505-seg6-mobile-v2-0-...."
Without this server, the same questions either require a working
~/.pwclientrc (which has trouble with CI checks and comment threads)
or hand-rolled curl against the REST API.
Install
Requires Python 3.10+. Install with uv:
# From PyPI (preferred once published)
uv tool install patchwork-mcp-server
# From a local checkout
uv tool install .
# From a Git remote
uv tool install git+https://github.com/higebu/patchwork-mcp-serverThe package installs a patchwork-mcp-server executable into ~/.local/bin/. To
refresh an existing install after editing source locally:
uv tool install --reinstall .Related MCP server: Gerrit MCP Server
Configuration
Variable | Default | Purpose |
|
| Base URL of the Patchwork instance |
|
| REST API version segment |
Read-only access to public Patchwork instances needs no credentials.
Connect to Claude Code
After uv tool install, the executable is on your PATH, so registration
needs no path:
claude mcp add patchwork-mcp-server -- patchwork-mcp-serverFor Claude Desktop, add it to ~/.config/claude/claude_desktop_config.json:
{
"mcpServers": {
"patchwork": {
"command": "patchwork-mcp-server"
}
}
}Tools
Tool | Purpose |
| List Patchwork projects on the instance. |
| Look up patches/covers by Message-Id (works across projects). |
| Series with every patch's state, delegate, archived flag, check summary. |
| Single patch with submitter, tags, series links, check summary. |
| CI checks for a patch (checkpatch, build_*, sashiko-*, contest, ...). Optional |
| Comment thread on a cover letter or patch (time-ordered, body preview). |
| Recent cover letters on a project, optional submitter filter, optional revision collapse. |
| Substring search over patch names, with project/state/submitter filters. |
Example interaction
What's the state of my SRv6 Mobile series on netdev?
Behind the scenes the agent calls:
find_by_msgid("20260505-seg6-mobile-v2-0-9e8022bdfdb6@gmail.com")to discover series id1089407onnetdevbpf.get_series(1089407)for the per-patch state table.get_checks(<lead_patch>, fail_only=True)to triage CI.get_comments("cover", 14117478)to read the thread.
Development
cd patchwork-mcp-server
uv sync --group dev
just check # ruff lint + format check + pytestjust release patch bumps the version, regenerates CHANGELOG.md via
git-cliff, and tags locally. Push the tag to trigger the PyPI release
workflow.
License
MIT
Available Tools
8 toolsfind_by_msgidA
Look up patches and cover letters by Message-Id across all projects.
A single Message-Id can appear under multiple projects (e.g. netdevbpf and linux-kselftest). Use this to discover the projects + series ids before drilling in.
Inputs: msgid (required).
get_checksA
Get CI check results for a patch (checkpatch, build_*, sashiko-*, contest, ...).
Pair with get_patch's check_summary to triage failing series quickly.
Inputs: patch_id (required), fail_only. Returns: result
get_commentsB
Get the comment thread on a cover letter or a patch, time-ordered.
Inputs: kind (required), id (required), preview_lines. Returns: result
get_patchC
Get a single patch with submitter, delegate, state, tags, and check summary.
Inputs: patch_id (required).
get_seriesA
Get a series with every patch's state, delegate, and archived flag.
Use this after find_by_msgid to see the whole patchset at once.
Inputs: series_id (required).
list_projectsA
List Patchwork projects available on the configured instance.
Returns id, link_name (used as ?project= filter), display name, and list_id.
Inputs: no named parameters. Returns: result
recent_seriesC
List recent cover letters on a project, optionally filtered by submitter.
With collapse_revisions=True, older revisions of the same series name are dropped so you only see the latest version of each topic.
Inputs: project (required), submitter, limit, collapse_revisions. Returns: result
search_patchesA
Search patches by name substring with optional project/state/submitter filters.
Inputs: query (required), project, state, submitter, +2 more. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 8 tools
Each tool targets a distinct operation: lookup by message ID, get patch details, get series, get comments, get checks, list projects, list recent series, and search patches. No two tools have overlapping purposes.
Tools mostly follow a verb_object pattern with underscores (e.g., get_patch, list_projects). 'recent_series' deviates slightly as adjective_object, but overall style is consistent and clear.
With 8 tools, the server covers core interactions without being too sparse or overwhelming. Each tool serves a clear need within the Patchwork domain.
The set covers essential read operations: project listing, patch/series retrieval, comments, checks, and search. Missing creation or modification tools, but that aligns with a read-focused MCP server.
Maintenance
Related MCP Connectors
Text-to-SQL MCP server: read-only queries on PostgreSQL, MySQL and SQL Server with your real schema
Agent-native MCP server over the public saagarpatel.dev corpus. Read-only, stateless.
The Cortex MCP server provides read-only access to real-time engineering context from the Cortex developer portal, allowing AI coding assistants to answer natural language questions about your organization's catalog (microservices, libraries, domains, teams, infrastructure), scorecards (engineering standards and best practices), initiatives (goals and deadlines), and Engineering Intelligence metrics. It includes tools for querying documentation, tracking personal entities, and accessing AI-assisted insights across the entire Cortex ecosystem.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceMCP server for OSIDB (Red Hat Product Security's vulnerability management system). Enables read-only querying of flaws, affects, trackers, and status via natural language from MCP clients like Cursor or Claude Desktop.2MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server that enables language models to query and manage Gerrit code reviews via the REST API.78 PyPI1Apache 2.0
- AlicenseNot gradedqualityFmaintenanceA read-only MCP server that exposes Quickwit log search and aggregations to LLM clients, enabling natural language log investigation.Apache 2.0
- AlicenseNot gradedqualityCmaintenanceMCP server enabling structured search and querying over Linux kernel mailing list archives (lore.kernel.org) for LLM-powered developer tools.148 PyPI5MIT