String

Markdown that runs. One file, any agent.

The web gave humans a universal surface for information. String aims to give AI agents the same — for work. A .sfmd file is a markdown document an agent can both read (like Markdown) and execute (calls APIs, runs CLI tools, navigates pages). Same file works in any runtime that speaks the format. No SDK. No registry. No per-agent code.

See it

A complete app, in one file:

---
name: weather
type: app
default: now
---

# Weather

Get current conditions anywhere on Earth.

```act.now
GET https://wttr.in/{city}?format=%l:+%C+%t+%w&m
  city, -c: string (required) "City name"
```

Install and call:

npm install -g @string-os/string
string '/install --app ./weather.md'
string app:weather '/act.now Seoul'
# → Seoul: ☀️ +20°C ↘6km/h

A markdown file declared the API. The runtime called it. The agent got the result. No protocol handshake, no per-tool server, no HTML parsing.

What it is

Format. SFMD (String Flavored Markdown) is 100% CommonMark with a few inline conventions: [!nav:main] for navigation menus, [@slug] for shortcut references, and fenced ```act.<id> blocks that declare typed callable actions (HTTP, CLI, or anything else the runtime supports).

Runtime. This repo ships @string-os/string — Browser, Session, Loader, Resolver. It loads SFMD files from file://, http(s)://, or installed packages; exposes their actions as /act.<name> calls; tracks per-session state; and renders results as Markdown.

What's standardized. Two verbs cover navigation and execution:

• /open — see something (document, page, app, URL, shortcut)
• /act — do something (call an API, run a CLI tool, submit data)

Different resource types get the same shape:

The agent learns the verbs once and uses them everywhere. New capabilities come from new documents, not new code.

Try it

npm install -g @string-os/string
git clone https://github.com/string-os/cookbook.git
cd cookbook

string setup '/install --app ./apps/weather/string.md'
string app:weather '/act.now Seoul'

The cookbook has a dozen runnable examples — kanban over GitHub Projects, an AI social network, search, code review, k8s helpers — each a single .sfmd file you can read end-to-end.

Other ways to embed the runtime

• MCP server (Claude Desktop, Cursor, …) — add @string-os/string-mcp to your MCP config
• In-process library — import { Browser } from '@string-os/string'. No daemon, no HTTP
• HTTP daemon + any-language client — string --daemon start. Wire spec in stringd protocol v0.1; reference TS client in @string-os/client

Add a feature once, it works in all four paths.

How it compares

• vs MCP. MCP is a protocol with a custom server per tool. SFMD is a file — works over file://, HTTP, email, or a USB drive. An SFMD file can be served by an MCP server (that's what string-mcp does), but it isn't tied to one.
• vs llms.txt. llms.txt is a static index, read-only. SFMD is read and execute. Actions are first-class.
• vs SKILL.md / agent skills. A SKILL.md is written for one agent runtime. An SFMD file is an app surface any runtime can read and execute.

The novelty is the combination: human-readable + AI-readable without HTML parsing + AI-executable + cross-agent portable, in one Markdown primitive with a small runtime.

v0.1 — what ships now

Working:

• SFMD parser, runtime, CLI, MCP server, daemon
• App / tool install from local files or HTTPS
• Action methods: GET, POST, PUT, PATCH, DELETE, CLI
• bash: topics and /exec (opt-in)
• Cookbook with a dozen working apps

Not yet:

• Signed packages — run SFMD files from trusted sources only
• Fine-grained capability permissions (default allowlist is restrictive)

Platform. Tested on Linux. macOS should work — /bin/bash is available and CLI actions use POSIX shell features only — but isn't routinely tested yet. Windows is not supported in 0.1.x: the runtime spawns /bin/bash for every CLI action. Use WSL on Windows, or wait for portable execution in v0.2.

Trust model in SECURITY.md. The full spec for parser implementors lives in the SFMD spec repo.

Packages

More

• docs.string-os.org — full guide
• Cookbook — runnable example apps
• SFMD spec — format specification
• Skill for AI agents — written for agent self-onboarding

Contributing: CONTRIBUTING.md. Security: SECURITY.md. License: MIT.