Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
제작자가 쓴 설명(영어 원문)
Inspect files, extract content and make guarded edits locally. Linux x86_64, glibc 2.35+.
이 세 가지는 절대 하나의 점수로 합치지 않아요. 의견, 측정, 인기는 서로 다른 것이니까요.
이 서버는 아직 쉬운 말로 된 요약을 쓰지 못했어요. 아래는 레지스트리에서 읽어 온 사실이에요.
쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.
어떤 앱을 쓰세요?
위에서 앱을 고르세요. 기억해 두었다가 다음부터는 버튼에 “…에 설치”라고 표시해요.
링크는 제작자의 파일로 바로 연결돼요. 우리가 보관하거나 다시 패키징하지 않아요. 파일을 열면 Claude Desktop이 설치해요.
터미널에서 이 명령어를 실행하세요
브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
이 서버는 내 컴퓨터에서 실행되는데, ChatGPT는 인터넷에 있는 서버에만 연결할 수 있어요. 다른 앱을 고르세요.
선택한 앱은 이 기기의 쿠키에만 저장돼요. 계정은 필요 없어요.
이 설정은 제작자의 레지스트리 정보로 만들었어요. README와 다르면 README를 따르세요.
내 컴퓨터에서 실행돼요. 설치하면 AI 앱에 도구 목록이 나타나요.
누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 6일 점검
레지스트리 이름이 GitHub 계정 krauqllc 기준으로 확인되었고, 이 계정이 저장소를 소유하고 있어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
내 컴퓨터에서 실행되는 서버라 인터넷 엔드포인트가 없어요
내 컴퓨터에서 실행되며, 대신 로컬 설정을 사용해요
내 컴퓨터에서 실행돼요. 도구 목록은 인터넷 엔드포인트가 있는 서버에서만 읽을 수 있어요
레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.
해당하는 기본 점검 1개 중 1개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.
레지스트리에 필수 키가 나와 있지 않아요. 다만 처음 쓸 때 로그인을 요구하는 서버도 있어요.
내 컴퓨터에서 실행돼요. AI 앱이 필요할 때 실행하고, 데이터는 제작자의 서버를 거치지 않아요(서버 자체가 온라인 서비스를 호출하는 경우는 제외).
GitHub 저장소는 마지막으로 – 변경됐고, 최신 버전(4.0.0-1)은 4일 전 레지스트리에 올라왔어요.
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. 내 컴퓨터에 MCPB도 있어야 해요.
같은 카테고리에는: Spec Driven Development, Remote Desktop Commander, Fomu MCP
같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.
카드 보는 법:★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Kenya government form guides: checklists, draft letters, eCitizen, Huduma, timelines. Sample data.
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
One local tool for every file an AI agent touches. Find what is hidden or risky in incoming files, including text written to instruct an AI, read them, change them without breaking them, and hand back clean copies. One offline CLI and local MCP server for Linux, with a JSON receipt for every step.
ufo inspect --deep reports comments, tracked changes, hidden text, hidden sheets, rows and slides, speaker notes, document properties, macros and embedded files in supported formats. Its text scan flags lines that read as instructions to an AI assistant, personal-data and credential patterns and look-alike mixed-script words, and it names a program travelling under a document's name. ufo inspect --human prints the same findings readably.ufo text extracts content with provenance, and --format structure adds addresses and hashes. ufo capabilities --file tells an agent what one file allows: bounded reads, candidate edits and their exact help, or why it should stop.ufo text --format structure --targets derives exact guard templates for the change you choose.ufo clean writes a copy without the removable hidden content the inspection found, and its receipt lists what was removed and what was not.Inspection reports its coverage gaps. Basic information for an unknown file does not imply a dedicated reader, editor or security inspector for its format, and findings are advisory: UFO is not antivirus or a guarantee that content is safe. File and resource limits apply.
Free for individuals and small teams. Every command and supported file type is included, with no signup or watermark. Organizations with fewer than ten people and under USD 1 million in annual revenue, including funding, qualify; any organization can evaluate it free. See the eligibility details.
Website and comparison with other agent tools: https://universalfileopener.com/cli/
Native Linux x86_64 installation needs glibc 2.35 or later (Ubuntu 24.04 and Debian 12 tested). Native Alpine/musl installation is unsupported. The Java runtime is included; the container option below uses its own runtime.
curl -fsSL https://universalfileopener.com/install.sh | sh
ufo doctor ufo skill install --agent claude-code --dir "$PWD"
install.sh downloads the app image from the latest release, checks it against the release's SHA256SUMS and installs it under ~/.local, without sudo. The same release has a Debian package (sudo apt install ./universal-file-opener_4.0.0_amd64.deb) and ufo-cli-4.0.0-docker.tar, an archive of the container image for offline transfer.
From a directory containing report.docx, download and run the public image:
docker run --rm --network none --read-only --cap-drop ALL \
--user "$(id -u):$(id -g)" --security-opt no-new-privileges \ --pids-limit 128 --cpus 2 --memory 1g \ --tmpfs /tmp:rw,nosuid,nodev,size=256m \ --mount type=bind,src="$PWD",dst=/input,readonly \ ghcr.io/krauqllc/ufo-cli:4.0.0 text /input/report.docx
No GitHub login is needed. The temporary mount is required. For edits, mount a separate writable output folder at /output and select a new file there. The container reads, edits and recognizes text; page rendering and Office-to-PDF conversion need the application package. For a deployment pinned to an immutable image, use the digest in the release notes.
Run in the project folder. For Cursor use ufo skill install --agent cursor --dir "$PWD". For Codex use ufo skill install --agent codex --dir "$PWD" --replace; it adds or updates only the marked UFO section of the project's AGENTS.md, keeping the rest. Start a new session after installing the guide.
For an MCP client instead of the skill, ufo mcp print-config --client claude-code prints a template. Replace its executable and folder placeholders, then merge it into the client configuration. Printing does not register a server. Follow the complete setup instructions.
From a clone of this repository, with Python 3.10+ and jsonschema installed:
python3 tools/cli/first_run_demo.py --ufo "$(command -v ufo)" \
--expect-version 4.0.0 --output first-look --mcp
The short example generates two fictional inputs. It collects basic information for an unfamiliar file, finds document properties, a reviewer comment and hidden text in a supported DOCX, and writes a clean copy. It independently checks the saved ZIP/XML bytes, original hashes, full receipt schemas and refusal to replace an existing output. Open first-look/START_HERE.txt to review the results.
Every notInspected gap and skipped clean category remains in the evidence. The optional MCP check discovers tools and calls inspection over local stdio; it does not establish that an installed assistant has connected successfully.
See a complete mixed-file editing workflow
Watch the 64-second film, then inspect and replay the workflow. A fictional change brief updates a spreadsheet model, a Word memo with native tracked revisions, and two slides in a board deck.
The film uses actual Microsoft Office exports of the downloadable files. The replay needs only Python 3 and UFO, with no agent or model API. It checks all 20 recomputed saved results, existing formulas, earlier periods, native Word revisions, original hashes and untouched package parts. Every edit matches its dry-run output hash; an intentionally stale source hash refuses. This is an edited script demonstration, not an agent recording or a timing benchmark.
Read the help for the operation you need: ufo edit paragraphs --help. Locate text with ufo find --text 'budget' --compact report.docx, then read selected paragraphs, cells or slides with ufo text --format structure --compact. Batch independent changes to one file, and reuse the output receipt's hash for the next edit. Store full receipts when you need an audit record.
--compact applies to text, find, inspect --deep and edit; intake, convert and unpack do not accept it. For MCP, read tools/list schemas: find takes one file and report requires an output path.
Every edit operation accepts --expect-sha256 (MCP options.expectedSha256), so a stale source refuses instead of being edited. Check each operation's help for its other flags rather than copying another operation's. For an unfamiliar file, start with ufo capabilities --file FILE. Always review what a receipt skipped and inspect the output before handoff.
The binaries are on the Releases page. This repository holds what you need to check them:
| Path | What it is |
|---|---|
docs/cli/schemas/ | JSON schemas for every receipt and report the CLI writes |
tools/cli/first_run_demo.py | The short unfamiliar-file and inspection/clean-copy example, with independent checks |
tools/cli/own_file_evaluation.py | Runs UFO on your own files: reads, reversible edits and an independent check of every output |
tools/cli/evaluate_sample.py | The end-to-end evaluation on generated sample files |
tools/corpus/compare_agent_tools.py | The comparison with other agent file tools, on the same tasks and checks |
tools/corpus/realworld_corpus_manifest.tsv | The public files that comparison uses, with their sources and SHA-256 |
tools/corpus/download_realworld_corpus.py | Downloads those files and verifies each one |
The other files under tools/ are the helpers these import. They need Python 3.10 or newer, jsonschema for receipt validation, and an installed UFO build.
Reviewed distribution source is also available under packaging/npm/, packaging/linux-cli/registry/ and packaging/homebrew/. These wrappers and descriptors target the same Linux x86_64 release. The Homebrew tap is public. Homebrew 7 loads formulae from other taps only after you trust them: brew trust --formula krauqllc/ufo/ufo, then brew install krauqllc/ufo/ufo. The release also provides a complete local MCPB bundle for compatible Linux clients, listed in the official MCP Registry. Directory revision 4.0.0-1 provides the matching 4.0.0 engine and bundle. Choose only existing task folders when installing the bundle; no directories are granted by default. The npm package @krauq/ufo-cli installs the same verified runtime; version 4.0.0 appears there once the publisher approves it with 2FA, which npm view @krauq/ufo-cli versions shows. On npm 12 use npm install --global @krauq/ufo-cli@4.0.0 --allow-scripts=@krauq/ufo-cli; on npm 11 or earlier, omit --allow-scripts and keep installation scripts enabled. The wrapper verifies and installs this exact Linux runtime. The server is also listed on Glama.
UFO=$(command -v ufo)
python3 tools/cli/evaluate_sample.py --ufo "$UFO" --expect-version 4.0.0 --output sample-result python3 tools/cli/own_file_evaluation.py --files ~/my-files --ufo "$UFO" --expect-version 4.0.0 --output my-files-report python3 tools/corpus/download_realworld_corpus.py --download --output corpus python3 tools/corpus/compare_agent_tools.py --download --corpus tools/corpus/realworld_corpus_manifest.tsv --corpus-root corpus --ufo "$UFO"
--download fetches the pinned releases of the other tools, and --setup-libraries creates the Python library environment the comparison uses. Both need the network once; the runs themselves do not.
The files here are Apache 2.0 (LICENSE, NOTICE). The binaries are licensed under ENGINE-LICENSE.txt: free for individuals, evaluation, education, non-commercial use, and organizations with fewer than ten people and under USD 1 million in annual revenue, including funding. Larger organizations need a license.
Install a delivered file with ufo license install ./license.json, then run ufo license status. A fresh file arrives for each paid period; install renewals with ufo license install ./license.json --replace. Checks are offline. For delivery or recovery help, contact support@krauq.com with the order number.
Report a security problem to support@krauq.com with "Security" in the subject.