# Cvx > CVX — structured input, professional output. YAML in, designed CV PDFs out, entirely on your machine. - **Type:** MCP server - **Install:** `agentstack add mcp-hrtips-cvx` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [hrtips](https://agentstack.voostack.com/s/hrtips) - **Installs:** 0 - **Category:** [Content & Media](https://agentstack.voostack.com/c/content-and-media) - **Latest version:** 0.1.0 - **License:** Apache-2.0 - **Upstream author:** [hrtips](https://github.com/hrtips) - **Source:** https://github.com/hrtips/cvx ## Install ```sh agentstack add mcp-hrtips-cvx ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About Structured input. Professional output. To the power of X. **CVX turns plain YAML into a professionally designed CV PDF — one command, fully local, no accounts; your data never leaves your machine.** Use it two ways: edit the YAML yourself, or have any AI assistant write it for you from your existing CV — CVX validates the files and tells the assistant exactly what to fix. > **🤖 AI assistants:** if a user asked you to create their CV with CVX, follow the default flow in the [AI guide](https://raw.githubusercontent.com/hrtips/cvx/main/docs/ai-guide.md) (section "Default flow (for assistants)"). Non-negotiables even if you skip the link: **never invent facts** — every entry must be truthful to the user's input; **linkedin.com URLs are unfetchable even when public** — don't infer, ask for the profile's **More → Save to PDF** export or pasted text instead; ask for a **square photo** in your first reply (it can't be generated; the CV renders fine without one); and CVX is the only renderer — never substitute another PDF generator. **No terminal? No problem.** Paste this into any AI chat (Claude, ChatGPT, Gemini, …): ```text Create my CV with https://github.com/hrtips/cvx — open that page and follow its instructions for AI assistants. Here is my LinkedIn profile: (or attach your current CV and a photo instead). ``` Page 1 of the scaffolded example CV — plus an ATS-safe single-column variant via --ats. Pagination has improved since these were captured; your build may split pages differently. --- ## Create your CV in two minutes ```bash npx @hrtips/cvx init # scaffold cv-content/ with a complete example CV npx @hrtips/cvx build # render it to a PDF ``` `init` gives you a finished, working CV — **Bruce Wayne's**, and yes, really. Open `bruce-wayne.pdf` and you're looking at a designed two-page CV: photo, sidebar, achievements, the lot. Now make it yours. Open the `cv-content/` folder, and replace Bruce's details with your own, one file at a time: ``` cv-content/ personal.yaml ← start here: your name, title, contact details summary.yaml ← the bullet points at the top of page 1 experience.yaml ← your work history (the bulk of the CV) education.yaml ← degrees, institutions, years competencies.yaml ← skill pills in the sidebar achievements.yaml ← awards and recognitions referees.yaml ← referees, or [] for "available upon request" images/profile.jpg ← your photo (square, 400×400px or larger) ``` Re-run `npx @hrtips/cvx build` after each file and watch the PDF update — seeing Bruce's entry next to yours makes the format self-explanatory. The output file is named after you automatically (`jane-doe.pdf`). Made a typo or unsure a file is right? `npx @hrtips/cvx validate` checks everything at once and tells you exactly what to fix: ``` cv-content/personal.yaml ⚠ unknown key "linkdin" ↳ did you mean "linkedin"? ``` Every scaffolded file also carries a `$schema` header, so editors with YAML support (VS Code + the YAML extension, JetBrains, …) autocomplete keys and flag mistakes as you type. Applying through a job portal? Generate the ATS-safe variant too — single column, no colours, machine-friendly: ```bash npx @hrtips/cvx build --ats ``` ### CLI reference | Command | Does | |---|---| | `npx @hrtips/cvx init` | Scaffold `cv-content/` with the example CV (won't overwrite an existing one) | | `npx @hrtips/cvx validate` | Check `cv-content/` — every problem at once, with file + field paths and fixes | | `npx @hrtips/cvx validate --strict` | Also fail on warnings (unknown keys); recommended for agents/CI | | `npx @hrtips/cvx build` | Render `cv-content/` to `.pdf` | | `npx @hrtips/cvx build --ats` | Render the ATS-safe single-column variant | | `npx @hrtips/cvx list` | Show available themes and layouts | | `npx @hrtips/cvx --help` / `--version` | Help / version | All commands accept `--json` for machine-readable output (one JSON object on stdout, logs on stderr) and use semantic exit codes: `0` ok, `2` validation failed, `3` render failed, `64` usage error. `init` is a convenience, not a prerequisite — `build` renders any `cv-content/` folder with valid YAML (built-in themes and layouts need no extra files). **Content schema:** content files are versioned by `schemaVersion` in `config.yaml` (currently `1`) and validated against the [canonical JSON Schema](schema/v1/cvx.schema.json). New keys may appear within a schema major. Keys can also be **removed** when they are measured to do nothing — `page1ExperienceCount` and `page1SplitBullets` were, in 1.8.0 — in which case builds keep working and `validate` names the removal, while `validate --strict` treats the now-unknown key as an error. The CHANGELOG says so at the release that does it. --- ## What goes in each file These are excerpts from the scaffolded example — build it once and you can see exactly where each snippet lands on the page. **personal.yaml** — the header and contact block: ```yaml name: Bruce Wayne title: Founder & Field Commander – Gotham Operations company: Wayne Enterprises phone: "+1 (201) 555-2283" phoneHref: "tel:+12015552283" email: bruce.wayne@wayne-enterprises.com linkedin: linkedin.com/in/brucewayne linkedinHref: "https://www.linkedin.com/in/brucewayne" ``` **experience.yaml** — one entry per role; `progression` (optional) renders a title history inside the entry: ```yaml - role: Founder & Field Commander – Gotham Operations company: The Batman period: 2005 – Present description: Self-directed vigilante operation safeguarding Gotham City through deterrence, investigation, and crisis response. progression: - title: Commander, Batman Incorporated period: 2011 – Present - title: Solo Operative, The Dark Knight period: 2005 – 2008 bullets: - Established and scaled a citywide security operation from a solo initiative to a franchised network (Batman Incorporated). - Recruited, trained, and led a high-performing field team including Robin, Nightwing, and Batgirl. - Reduced organised-crime activity in Gotham by an estimated 60% through data-driven surveillance and rapid incident response. ``` **summary.yaml** — the bullet list at the top of page 1: ```yaml - "Strategic operations leader with 20+ years' experience, progressing from solo field operative to Field Commander of a citywide security network." - "Co-founded the Justice League as a global response coalition, serving as chief strategist and contingency planner for existential-scale threats." ``` **education.yaml** ```yaml - degree: "Applied Sciences & Criminology (self-directed)" institution: League of Shadows period: 1998 – 2004 - degree: BSc, Criminology & Chemistry institution: Gotham University period: 1994 – 1998 ``` **competencies.yaml** — rendered as skill pills in the sidebar: ```yaml - Strategic Planning - Criminal Investigation - Crisis Response - Surveillance & Intelligence ``` **achievements.yaml** ```yaml - year: Gotham's Most Influential Citizen text: "— 2024, Gotham Gazette" - year: Key to the City text: "— Office of the Mayor, Gotham City" ``` **referees.yaml** (use `[]` to print "available upon request"): ```yaml - name: Diana Prince title: Founding Member, Justice League company: Themysciran Embassy email: d.prince@justiceleague.org phone: "+1 (202) 555-0177" ``` Delete what you don't need — an empty file (or `[]`) simply drops that section from the CV. Any new `.yaml` file you drop into `cv-content/` is auto-discovered as a content key. The complete field-by-field schema for every file lives in [docs/cv-schema.md](docs/cv-schema.md). ### Let an AI write the YAML for you The formats are deliberately LLM-friendly, and the schema is published for machines ([llms.txt](llms.txt), [docs/cv-schema.md](docs/cv-schema.md)). **[The AI guide](docs/ai-guide.md)** has copy-paste prompts for every route; the short version: **Coding agent** (Claude Code, Cursor, …) — lowest friction: run `npx @hrtips/cvx init`, then ask the agent to replace the example content with your details and build. The scaffolded `cv-content/README.md` documents the schema, so the agent edits, runs `npx @hrtips/cvx build`, and fixes errors itself. **Chat assistant** (Claude, ChatGPT, …) — paste your existing CV or LinkedIn profile text along with this prompt: > Read the CVX content schema at https://raw.githubusercontent.com/hrtips/cvx/main/docs/cv-schema.md then convert my CV below into CVX cv-content/ YAML files. Output each file in its own fenced code block titled with the filename. Keep every fact truthful to my input — don't invent anything. Save the generated files into `cv-content/`, drop in your photo, run `npx @hrtips/cvx build`. No web access in your assistant? Use the [self-contained prompt](docs/ai-guide.md#route-c--chat-assistant-self-contained-prompt). ### Plug it into your agent (MCP) CVX ships an MCP server — any MCP client (Claude Desktop, Claude Code, Cursor, VS Code, …) can drive the whole loop with five tools: `get_schema`, `init_cv`, `validate_cv`, `build_pdf`, and `plan_layout` (a dry run that reports how the CV paginates — page count, per-page column fills, what landed where — without writing a PDF). No API keys, fully offline. CVX renders 100% of your YAML and never drops, clips, or hides text to fit a page — so an assistant driving it can't quietly cut a section to hit a page count either. If the CV runs longer than you want, it surfaces the trade-off and you decide what goes. ```bash npx @hrtips/cvx mcp init --client claude # Claude Code (.mcp.json, project) npx @hrtips/cvx mcp init --client claude-desktop # Claude Desktop (global config) npx @hrtips/cvx mcp init --client cursor # Cursor (.cursor/mcp.json) npx @hrtips/cvx mcp init --client vscode # VS Code (.vscode/mcp.json) ``` Then restart the client and ask it to make your CV — it fetches the schema, scaffolds, fills in your details, validates after every edit, and renders the PDF. The config writer merges into existing files; it never clobbers other servers. There's also a ready-made [Agent Skill](skills/cvx/SKILL.md) with the same loop for skill-capable agents. ### Build it into a ChatGPT GPT ChatGPT's sandbox has a Node runtime but no network, so `npx` cannot work there. CVX therefore ships as **one self-contained file** — schema, template, fonts and all — attached to [every release](https://github.com/hrtips/cvx/releases/latest) as `cvx.bundle.min.js`. It needs nothing but Node 20+: no install, no `node_modules`, no network. Upload it to any ChatGPT conversation and ask it to run — or spend three minutes wiring a GPT that fetches the current release itself, so it never goes stale. **[docs/custom-gpt.md](docs/custom-gpt.md)** has both, including the instructions to paste, the action to import, and the [privacy policy](https://hrtips.github.io/cvx/privacy) URL a GPT with an action needs before it can be shared. The GPT does what a chat assistant otherwise cannot: it renders the PDF, **opens it and looks at the pages**, then fixes the layout before you ever see it. ### Your photo Drop it into `cv-content/images/` as `profile.` — `jpg`, `jpeg`, `png`, or `webp` are auto-detected (that order wins if several exist). Square crop, at least 400×400px. --- ## Themes, layouts, and page flow Everything visual is controlled by `cv-content/config.yaml`: ```yaml theme: teal # teal | coral | mono layout: two-column # two-column | single-column ``` Change a value, re-run `npx @hrtips/cvx build`, done. **Themes** control colour and styling: | Theme | Accent | Description | |---|---|---| | `teal` | `#1a6070` | Professional teal (default) | | `coral` | `#c0534a` | Warm coral red | | `mono` | `#000000` | Black and white, ATS-optimised | **Layouts** control page structure: | Layout | Structure | Description | |---|---|---| | `two-column` | Sidebar + main column | Designed CV with photo, identity block, achievements | | `single-column` | Full width | ATS-safe, no sidebar, no decorative elements | **Pagination** — experience entries are distributed across pages automatically (greedy bin-packing), never overflowing a page, and an entry too tall for the remaining room is split at a bullet boundary and continued overleaf. There are no pagination settings: the layout follows the content. (The old `page1ExperienceCount` / `page1SplitBullets` keys were removed — measured, they never reduced the page count, and forcing them pushed content onto an unnumbered extra sheet. A config that still has them gets a validation message saying exactly that.) ### Script support **CV rendering is English/Western-European Latin only.** CVX bundles [Lato](https://fonts.google.com/specimen/Lato) and registers no fallback font, so scripts Lato doesn't cover — Cyrillic, Greek, Vietnamese, Turkish `ş`/`ğ`, Czech/Romanian diacritics, and all non-Latin scripts (Devanagari, Tamil, Sinhala, CJK, Arabic, …) — render invisibly. `cvx validate` and `cvx build` warn loudly when your content contains characters the bundled font cannot draw, so this fails visibly rather than silently. This is a deliberate scope decision, not an oversight: shipping fallback fonts for those scripts would blow the package-size budget many times over. (The project website is multilingual; the renderer is not — the two are independent.) ### Custom layouts You can define your own page structure — drop a `.yaml` file into `cv-content/layouts/` and reference it by filename: ```yaml # cv-content/layouts/compact.yaml template: two-column pages: first: sidebar: - identity-photo - contact main: - summary - spacer: 27 - experience continuation: sidebar: - identity-compact - education - competencies - achievements main: - experience:continued last: sidebar: - identity-compact - referees main: - experience:continued ``` Then set `layout: compact` in `config.yaml`. **The sidebar's three lists are one ordered flow, not three page assignments.** `first.sidebar` + `continuation.sidebar` + `last.sidebar` are concatenated in that order, and CVX measures the result to decide which page each section lands on. `last.sidebar: [referees]` therefore means *"referees comes last in the sidebar"*, not *"referees renders on the last page"*. Use the buckets to express **order**; let pagination be measured. (`identity-photo`/`identity-compact` are the exception — they are injected at the top of every page's sidebar rather than packed.) The **main** lists remain per-page-kind: `first.main` on page 1, `last.main` on the final page, `continuation.main` in between. Available section keys: | Key | Renders | |---|---| | `identity-photo` | Name, title, company + profile photo (sidebar) | | `identity-compact` | Name, title, company without photo (sidebar) | | `contact` | Phone, email, LinkedIn, location with icons (sidebar) | | `achievements` | Year + description list (sidebar) | | `education` | Degree, institution, period (sidebar) | | `competencies` | Skill tags as pills (sidebar) | | `referees` | Referee contact details (sidebar) | | `summary` | Bullet point list (main column) | | `experience` | Experience entries for page 1 (main column) | | `experience:continued` | Continuation experience entries (main column) | | `header-ats` | Full-width name/title/contact header (single-column) | | `spacer: N` | Vertical spacer of N points | --- ## ATS & AI-parser keywords Both PDFs embed a keyword list into the standard **`Keywords` metadata field** — the field some applicant tracking systems (ATS) and AI CV parsers read. Keywords live in metadata, **not** as hidden text on the page. > **Reality check:** most mainstream ATS rank on text extracted from the CV *body*, and support for the PDF `Keywords`/XMP field is inconsistent. Treat this as a best-effort supplement to a keyword-rich b … ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [hrtips](https://github.com/hrtips) - **Source:** [hrtips/cvx](https://github.com/hrtips/cvx) - **License:** Apache-2.0 Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-hrtips-cvx - Seller: https://agentstack.voostack.com/s/hrtips - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.