Developers

Kami installs into an AI agent and runs in that environment. There is no hosted Kami API, account, or API key. Update checks and remote resources may use the network; see the privacy page for details.

The integration surface is therefore local: an MCP server for tool calls, JSON content schemas for structure, and CLI checks for verification. The static files listed at the bottom of this page are the machine-readable index of all of it, and they are generated from the same registry the templates use, with generation checks to catch drift.

# Claude Code, Codex, Cursor, and other agents
npx skills add tw93/kami -a claude-code codex cursor -g -y

# Claude Code plugin (v2.1.142+)
/plugin marketplace add tw93/kami
/plugin install kami@kami

# Codex plugin
codex plugin marketplace add tw93/kami
codex plugin add kami@kami

# MCP client, from a checkout
claude mcp add kami -- python3 <checkout>/skills/kami/scripts/mcp_server.py

Claude Desktop: download the release asset kami.zip, not GitHub's Source code ZIP, then upload it under Customize > Skills.

• kami_templates - list document templates, browser-only templates, the diagram library, and content schema types, with the reference docs to read before filling.
• kami_doctor - report installed PDF render, visual-check, editable-PPTX, and font capabilities without treating an unavailable engine as a clean result.
• kami_render - render trusted local Kami HTML to PDF through WeasyPrint, with build-time code highlighting. Returns the PDF path and page count.
• kami_check - run the deterministic checks for a file. Returns the readable report plus stable rule IDs, findings, engine coverage, and explicit degraded checks.
• kami_screenshot - rasterize every PDF page to PNG and return the paths, deterministic CJK font verdict, and a review_pending checklist contract for the perceptual pass.

The server speaks newline-delimited JSON-RPC 2.0 on stdin and stdout and has no third-party dependency for the protocol itself. Tools that need WeasyPrint, pypdf, or PyMuPDF surface the install hint as a tool error instead of crashing. Its card, including the protocol version and the tool list, is published at /.well-known/mcp/server-card.json.

Nine schemas live under references/schemas/: changelog, equity-report, landing-page, letter, long-doc, one-pager, portfolio, resume, slides. A new content.json also carries a brief with audience, job, output contract, target, preserve boundary, evidence, and acceptance checks; older IR files remain valid. Validate the content object before layout, then check how much of it actually reached the page:

python3 scripts/build.py --check-content content.json
python3 scripts/build.py --check-content content.json filled.html

The second form reports coverage: fields that exist in the content object but never made it into the document are the most common failure in agent-generated layouts, and can be easy to miss in a quick review.

• --check-placeholders - unfilled template text still in the document.
• --check-markdown - markdown syntax that leaked into the rendered output.
• --check-orphans - single lines stranded at a page break.
• --check-density - pages ending with more than a quarter of the page empty.
• --check-rhythm - slide sequences that repeat the same layout too many times.
• --check-resume-balance - column and section balance in a resume.
• --check-visual - exports page images and returns a fixed review checklist for a perceptual pass; the deterministic checks cannot see a page that is merely ugly.

Run python3 scripts/build.py --help for the authoritative list. Repository-wide checks, python3 scripts/build.py --check, cover template lint, design-token sync, and public-site fact drift.

• /llms.txt - product summary, install commands, and when an agent should reach for Kami.
• /index.md - the homepage as Markdown. A request to / carrying Accept: text/markdown, or /?mode=agent, redirects here.
• /developers.md - this page as Markdown.
• /developers/llms.txt - the developer surface on its own, for agents that only need the integration facts.
• /.well-known/agent-skills/index.json - Agent Skills discovery index, with a SHA-256 digest of the skill file it points to.
• /.well-known/mcp/server-card.json - MCP server card: version, protocol version, transport, and tool list.
• /feeds/catalog.jsonld - JSON-LD catalog of all 8 document templates and 18 diagram types, indexed by /schemamap.xml.
• /SKILL.md - the skill definition itself, served as text/markdown.
• /sitemap.xml and /robots.txt - crawl surface and the AI crawler policy.