CLI tools with XML/JSON output for AI agents. Built with AI tools to power AI tools.
When an AI runs ls it gets plaintext like:
-rw-r--r-- 1 user staff 1024 Apr 6 10:00 main.go
The agent must guess which column is size, date, filename, and so on.
$ aict ls src/
<ls timestamp="1234567890" total_entries="2">
<file name="main.go" size_bytes="1024" language="go" mime="text/x-go"/>
<file name="utils.go" size_bytes="512" language="go" mime="text/x-go"/>
</ls>
Every field is labeled. Paths are absolute. Timestamps are Unix integers. Language and MIME type are detected automatically.
go install github.com/synseqack/aict@latest
go install github.com/synseqack/aict/cmd/mcp@latest
Or build from source:
git clone https://github.com/synseqack/aict
cd aict
go build -o aict .
go build -o aict-mcp ./cmd/mcp
# XML (default)
aict ls src/
aict grep "func" . -r
# JSON
aict ls src/ --json
# Plain text
aict ls src/ --plain
| Category | Tools |
|---|---|
| File inspection | cat, head, tail, file, stat, wc |
| Search | ls, find, grep, diff |
| Path | realpath, basename, dirname, pwd |
| Text | sort, uniq, cut, tr |
| System | env, system, ps, df, du, checksums |
Build: go build -o aict-mcp ./cmd/mcp
Add to Claude Desktop (~/.config/claude/claude_desktop_config.json):
{
"mcpServers": {
"aict": {
"command": "aict-mcp",
"args": []
}
}
}For Claude Code, add to your ~/.claude.json:
{
"mcpServers": {
"aict": {
"command": "aict-mcp",
"args": []
}
}
}Or set the AICT_BINARY environment variable if the binary is not in PATH:
{
"mcpServers": {
"aict": {
"command": "aict-mcp",
"env": {
"AICT_BINARY": "/full/path/to/aict"
}
}
}
}User prompt: "Show me all files in the src directory"
Expected behavior:
- Run
aict ls src/ - Returns XML with file entries, each containing name, size, language, and MIME type
- Example output:
<ls timestamp="..." total_entries="2">
<file name="main.go" size_bytes="1024" language="go" mime="text/x-go"/>
<file name="utils.go" size_bytes="512" language="go" mime="text/x-go"/>
</ls>User prompt: "Find all files containing the word 'func' in the project"
Expected behavior:
- Run
aict grep "func" . -r - Returns XML with matching files, line numbers, and context
- Each match includes the line content and byte offset
User prompt: "Show me detailed info about main.go"
Expected behavior:
- Run
aict stat main.go - Returns XML with permissions, timestamps, owner, size, language, and MIME type
aict is read-only. It does not collect, store, or transmit any user data. For complete privacy information, see PRIVACY.md.
- No data collection
- No network requests (except local MIME type detection)
- No telemetry or analytics
- Only accesses paths explicitly provided
Why not pipe to jq?
jq doesn't help with ls, cat, find, stat - they don't output JSON. aict gives structured output for every tool.
Why XML by default?
Denser encoding for AI context windows. <file size="1024"/> is 22 chars vs 30+ in JSON. Use --json if preferred.
How is this different from ripgrep?
ripgrep is excellent for searching. aict grep provides similar functionality but adds language detection, MIME type, and integrates with the same output format as other tools. For pure grep performance, use ripgrep with --json and parse the output.
How is this different from eza/lsd?
eza and lsd are modern ls replacements with better colors and formatting. aict outputs structured data instead of human-readable tables. Use eza for terminal use, aict for AI agent consumption.
Does it work on Windows?
Partially. ls, cat, stat, wc, find, diff, grep, head, tail work. ps and df are Linux/macOS only.
Can I contribute?
Yes. See CONTRIBUTING.md for guidelines.
For issues and questions: https://github.com/synseqack/aict/issues
| Tool | GNU | aict | Notes |
|---|---|---|---|
| ls (1000 files) | ~2ms | ~15ms | 7x overhead |
| grep (100k lines) | ~1ms | ~100ms | language detection |
| cat (100k lines) | ~1ms | ~23ms | 17x overhead |
Use --plain to skip enrichment when you only need content.
MIT
