agent-atlas-cli

Agent Atlas

Search and research the open web for your AI agent.

Agent Atlas is a capability layer: it installs, health-checks, and routes free upstream tools so your AI coding agent can research the web. It is not a scraper framework — after setup, the agent calls upstream CLIs/APIs directly.

10 Western channels. No China-only platforms. English-first docs · Türkçe.

Not the same as other GitHub projects named “AgentAtlas” (e.g. browser schema registries). This repo is an open-web research installer + router for AI coding agents.

---

What it does

---

Channels & tools

Tier 0 — zero config (after install)

Tier 1 — login / session

Full command examples: docs/platforms.md · Tier 1 setup: docs/tier1.md

---

Quick start

Tell your AI agent:

Install Agent Atlas: https://raw.githubusercontent.com/batu3384/agent-atlas/main/docs/install.md

Or manually:

uv tool install agent-atlas-cli
# fallback: uv tool install git+https://github.com/batu3384/agent-atlas.git
# ensure ~/.local/bin is on PATH
agent-atlas install
agent-atlas doctor
agent-atlas smoke

From a clone:

git clone https://github.com/batu3384/agent-atlas.git
cd agent-atlas
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
agent-atlas install
agent-atlas doctor

Tier 1 (optional):

agent-atlas install --channels twitter,reddit,linkedin,opencli
agent-atlas configure twitter_chrome_profile "Profile 3"
agent-atlas configure opencli_profile atlas

Ensure ~/.local/bin is on your PATH if you use uv tool install for agent-atlas-cli or rdt-cli.

---

CLI

agent-atlas doctor          # human report
agent-atlas doctor --json   # for agents — use active_backend per channel
agent-atlas doctor --strict # exit 1 if Tier 0 off/error or any error
agent-atlas smoke           # one real call per ready channel
agent-atlas smoke --json
agent-atlas watch           # health + update; exit 1 = Tier-0 broken or update
agent-atlas check-update
agent-atlas sync-reddit     # opt-in Reddit cookie sync from Atlas Chrome
agent-atlas route URL       # which channel owns this URL
agent-atlas install         # Tier 0 + SKILL.md
agent-atlas install --channels twitter,reddit,linkedin,opencli,all
agent-atlas install --safe  # print needs only
agent-atlas configure KEY VALUE
agent-atlas skill --install
agent-atlas uninstall

---

Example research (after doctor --json)

# Prefer doctor JSON for routing
agent-atlas doctor --json
# → use each channel's active_backend

# Tier 0
curl -s "https://r.jina.ai/https://example.com"
mcporter call 'exa.web_search_exa(query: "AI agents 2026", numResults: 5)'
gh search repos "AI agent" --sort stars --limit 5
yt-dlp --flat-playlist --print "%(title)s" "ytsearch3:AI agents"
python3 -c "import feedparser; print([e.title for e in feedparser.parse('https://hnrss.org/frontpage').entries[:5]])"

# Tier 1
twitter search "AI agents" -n 10
rdt search "AI agents" -n 10 --compact --yaml
# LinkedIn: agent MCP tools after linkedin-scraper-mcp login
curl -s "https://r.jina.ai/https://www.linkedin.com/in/…"

---

Design

• Config: ~/.agent-atlas/config.yaml (mode 600)
• Temp: /tmp/ — do not pollute the project tree
• Skill: agentskills.io SKILL.md → ~/.agents/skills/agent-atlas/
• Doctor statuses: ok = ready · warn = installed / config incomplete or session not live-probed · off = missing/disabled · see docs/channels.md
• Inspiration: Agent Reach’s installer/doctor model — different product, English-first, Western-only scope

---

Docs

---

License

MIT — see LICENSE. Upstream tools keep their own licenses.

Credits (upstream)

Jina Reader · Exa · mcporter · yt-dlp · GitHub CLI · feedparser · OpenCLI · twitter-cli · rdt-cli · linkedin-scraper-mcp