walla

██╗    ██╗  █████╗  ██╗      ██╗       █████╗            ██████╗ ██╗      ██╗
██║    ██║ ██╔══██╗ ██║      ██║      ██╔══██╗          ██╔════╝ ██║      ██║
██║ █╗ ██║ ███████║ ██║      ██║      ███████║  █████╗  ██║      ██║      ██║
██║███╗██║ ██╔══██║ ██║      ██║      ██╔══██║  ╚════╝  ██║      ██║      ██║
╚███╔███╔╝ ██║  ██║ ███████╗ ███████╗ ██║  ██║          ╚██████╗ ███████╗ ██║
 ╚══╝╚══╝  ╚═╝  ╚═╝ ╚══════╝ ╚══════╝ ╚═╝  ╚═╝           ╚═════╝ ╚══════╝ ╚═╝
  look · offer · talk · sell
  Dr. Berend Gort  ·  www.berendgort.dev

▶ Play full demo with sound · Demo music: Funkee Monkeee by Michael Ramir C. (Mixkit Stock Music Free License).

walla turns Claude, GPT, or Cursor into a Wallapop helper that actually does the work for you. You drop a few photos and answer a couple of questions; the agent publishes the listing. When buyers message, it reads the chats, replies for you, and negotiates down to your minimum price — then shows you the best deal. You only open the Wallapop app to accept. No typing chats. No babysitting the inbox.

Unofficial Wallapop.es CLI + MCP + Python library. Not affiliated with Wallapop. Respect ToS and rate limits.

What this unlocks

You tell Cursor: “Find me a used kite near Barcelona under €400, message the seller, and offer if it’s a GRAB.”

The agent runs walla — ranks listings GRAB / LOOK / PASS, messages the shortlist, and follows replies until a price is agreed. You close the deal in the Wallapop app.

walla setup --lat 41.39 --lon 2.17 --km 30 --label Barcelona \
  --budget 400 --aggression fair --must "kite cabrinha"
walla login                         # paste session cookie once
walla pursue "tabla kite Cabrinha" --json
walla desk --json                   # one inbox read; replies; stack.best

Sell is the same idea: answer the photo questions, publish once, then desk talks to buyers.

walla sell ./shot.jpg --json
walla sell ./shot.jpg --title "…" --desc "…" --eur 40 \
  --category 10105 --root 12579 --floor 35 --yes --json

Quality bar (enforced in CI):

• docs/code_quality.md — Korotkevich / Tourist bar
• docs/data_engineering_standards.md — Gray / Stonebraker bar
• docs/WIRE.md — captured consumer API
• docs/rate_limit_analysis.md — polite spacing

For AI agents (mandatory protocol)

If the user pasted this GitHub link (or this README), follow this section exactly.

Goal

Look, offer, and talk on wallapop.es using the walla package. Speak briefly. Link listings as https://es.wallapop.com/item/<web_slug>. Verdicts: GRAB / LOOK / PASS.

Bootstrap

1. Install from PyPI (preferred):

   pipx install 'walla-cli[mcp]'
   

   From a clone only when hacking: pipx install -e '.[mcp]'.
2. walla doctor --json — read auth.session and network.reachable. Never print tokens or cookies.
3. If no session: ask the human to run walla login and paste __Secure-next-auth.session-token from es.wallapop.com (Cookie-Editor). Agents may pass it non-interactively: walla login --cookie '<value>' --json.
4. walla instruct --json — read mandate + hitl. If mandate.budget is null, ask once for budget / aggression / must words, then walla setup ….
5. Do the ask with walla … --json and narrate in prose.

Optional MCP:

{
  "mcpServers": {
    "walla": { "command": "walla-mcp" }
  }
}

Operating loop

HITL

Rules:

• pursue and desk send negotiation chat. Do not ask the human to send.
• Bare say / offer / sell / unsell still need --yes.
• --yes never completes a purchase or accepts a sale.
• If they only asked to look, do not send.
• Ambiguous match or multiple GRABs: ask once with candidates.
• Always present Wallapop shortlists ranked best-first with a markdown link on every item URL.
• Default search is all of Spain. Shipping counts anywhere. Use walla search --local only when the human asked for nearby or pickup.
• In-person offers outside pickup radius: refuse.
• Envelope: ok, api_version, data or error / error_type / retryable.

walla instruct --json returns this recipe.

Install

pipx install 'walla-cli[mcp]'   # PyPI: https://pypi.org/project/walla-cli/
walla setup --lat 41.39 --lon 2.17 --km 30 --label Barcelona --json
walla login          # paste __Secure-next-auth.session-token
walla doctor --json

CLI

Develop

pip install -e '.[dev,mcp]'
pytest -q -m 'not live'
pytest -m live          # hits Wallapop; needs network
ruff check .
mypy walla
python scripts/check_code_quality.py

Read AGENTS.md before extending.

Help improve walla

Long agent chats help. After you confirm, pick a PR method:

walla teach ./cursor-export.md --method a --yes --json   # A gh  B git  C browser  D hub  E manual

Cookies/tokens are stripped first. If gh/git/hub is missing, open https://github.com/berendgort/wallapop-cli/compare and file the PR by hand (parked file is in ~/.config/walla/teach/). Agents ask once on first use, then at 20 / 40 / 60 prompts — never on every search.

Disclaimer

Unofficial. Not affiliated with Wallapop. Personal / research use.

License

MIT