
WrapGuard MCP
io.github.Yuni9959Published by GitHub account Yuni9959v0.1.1Updated Oct 24, 2026
Detect unnatural Korean and English mid-word wrapping in responsive web pages.
Overview
Local MCP server that renders responsive pages in a real browser and reports Korean or English words split mid-word across lines.
- What you can do
- Audit pages for mid-word line breaks at several viewport widths, using real browser coordinates
- Detect hard source newlines embedded inside words and distinguish avoidable breaks from unavoidable long identifiers
- Inspect the wrapping CSS that applies and suggest CSS fixes, with dry-run previews
- Verify regressions and apply fixes only when writes are explicitly allowed
- Report server and version information
- When to use it
- Useful when you maintain Korean or English web typography and want an assistant to check responsive layouts for words broken across lines, or to gate a build on wrap regressions.
- Requirements
- Runs locally as an npm package via npx; needs Node.js 22 or newer and a Playwright Chromium installation (or an installed Chrome or Edge). Optional WRAPGUARD_CONFIG points to a config file and WRAPGUARD_SERVER_NAME sets the display name.
Installation
In SourceWeft
- Open WrapGuard MCP in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
WrapGuard MCP
WrapGuard is an MCP server and CLI that renders responsive pages with Playwright and finds Korean or English words split across visual lines. It also reports hard newlines embedded inside words, such as 업로↵드, without rewriting immutable source evidence.
한글·영어 문장이 반응형 화면에서 단어 중간에 잘리는 문제를 실제 브라우저 좌표로 검사합니다. 표시 이름은 자유롭게 바꿀 수 있지만 npm/GitHub/MCP Registry의 패키지 식별자는 별도로 유지됩니다.
What it checks
- Korean eojeol-like tokens and English word tokens, including apostrophes and hyphens
- Actual grapheme positions across 320, 375, 768, and 1024 px by default
- Avoidable mid-word wraps versus unavoidable long identifiers
- Hard source newlines between letters
- Applied wrapping CSS and same-origin matching rules
- Explicit, dry-run-first CSS suggestions
URLs, code, hashes, and immutable raw evidence can be excluded with patterns or data-wrapguard-ignore. Public fixtures contain no private data.
Install and run locally
Requirements: Node.js 22 or newer and a Playwright Chromium installation.
With browserChannel: "auto" (the default), WrapGuard tries Playwright Chromium first and then an installed Chrome or Microsoft Edge. Installing Playwright Chromium is the reproducible choice for CI.
After an npm release, an MCP client can launch it directly:
On Windows, clients that do not resolve command shims may need npx.cmd instead of npx.
Run it as a release gate without MCP:
Exit code 0 means no avoidable split, 1 means regression issues were found, and 2 means the audit could not run.
Change the displayed name
Copy wrapguard.config.example.json to wrapguard.config.json, then edit:
WRAPGUARD_SERVER_NAME overrides the JSON value. This changes the runtime MCP display name only. Before the first public release, change name, mcpName, repository URLs, and server.json together if you want a different permanent package identity.
MCP tools
server_infoaudit_pagedetect_midword_breaksinspect_wrap_csssuggest_wrap_fixverify_wrap_regressionapply_wrap_fix— dry-run by default; real writes requireallowWrites: trueand an allowed root
Safe defaults
The default configuration audits local files under the configured roots and localhost only. Remote URL auditing, screenshots, and CSS writes are disabled unless explicitly enabled. A hosted instance should use a narrow allowedHosts list because unrestricted browser navigation creates an SSRF risk.
CSS previews, CSS writes, and screenshots resolve filesystem links before checking allowedFileRoots. Screenshot directories must be inside an existing allowed root. An explicitly selected configuration file must exist and contain a JSON object; a missing conventional wrapguard.config.json still uses defaults. Runtime and MCP versions are read from package.json.
Three publication layers
- GitHub publishes source, issues, documentation, and releases. It does not by itself run an MCP endpoint.
- npm publishes the stdio executable so users can run
npx -y wrapguard-mcp. - Remote MCP hosts
/mcpover Streamable HTTP for clients that do not run local processes.
Publish source and npm
After reviewing the repository name and public contents:
Then validate and publish server.json with the official mcp-publisher. The npm package must be published first and the mcpName in package.json must match server.json.
Host a remote endpoint
The included container starts the Streamable HTTP server at /mcp:
For a public host, set WRAPGUARD_HTTP_HOSTS to the deployed domain and store WRAPGUARD_HTTP_TOKEN as a platform secret. Use a container service that supports Playwright/Chromium, such as Cloud Run, Fly.io, Render, Railway, or an equivalent service. Add the final HTTPS URL to the remotes field in server.json only after the endpoint exists.
Vercel Functions are not the default recommendation for this project because browser binaries, execution duration, and memory limits make browser audits less predictable than a long-running container.
Development
The test suite uses synthetic Korean and English pages and an in-memory MCP client/server connection.
License
MIT
Source: README.md at commit c065ed3
Tools
0Version history
1- v0.1.1LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts