olxcars

Search used cars on OLX India from a terminal, or plug it into an AI agent over MCP.

OLX has no public API. This tool talks to OLX's own search endpoint with a Chrome TLS fingerprint — plain curl/requests get connection-reset by OLX's edge.

Install

pip install -e .

Needs Python 3.10+. Dependencies: curl_cffi (transport) and mcp (server).

From GitHub, without cloning:

pip install "olxcars @ git+https://github.com/aswathas/olx-car-search"

As a Claude Code plugin

The repo is a plugin marketplace. One command installs the CLI, the MCP server and the aswath-olx-car-search skill:

claude plugin marketplace add aswathas/olx-car-search
claude plugin install olxcars@olx-car-search

The plugin runs the MCP server through uvx, so no Python environment is needed on the host.

Related MCP server: olx-india-mcp

CLI

olxcars search "swift" --city chennai --min-price 300000 --max-price 600000 \
    --year-min 2018 --year-max 2021 --max-km 40000 --fuel petrol \
    --transmission manual --owner first --sort price_asc --table

olxcars search "i20" --city bengaluru --max-price 800000 --json
olxcars cities chennai            # resolve a city name to an OLX location id
olxcars listing 1859805530        # full detail for one ad
olxcars mcp                       # run the MCP server on stdio

search prints JSON by default (agent-friendly); --table prints one line per car. --debug prints the outgoing request URL and OLX's own applied_filters echo on stderr — use it to confirm a filter actually landed (OLX ignores unknown params silently).

Each car comes back as:

{
  "id": "1859805530",
  "title": "Maruti Suzuki Swift LXI, 2018, Petrol",
  "price": 510000,
  "price_display": "₹ 5,10,000",
  "make": "Maruti Suzuki", "model": "Swift", "variant": "2018-2021 1.2 LXI",
  "year": 2018, "km": 61000,
  "fuel": "petrol", "transmission": "manual", "owner": "first",
  "city": "Chennai", "state": "Tamil Nadu", "lat": 13.045, "lon": 72.984,
  "seller": "individual", "posted": "2026-10-25",
  "images": ["https://..."],
  "url": "https://www.olx.in/item/1859805530/"
}

MCP server (for your AI agent)

claude mcp add olxcars -- olxcars mcp

Or in .mcp.json:

{ "mcpServers": { "olxcars": { "command": "olxcars", "args": ["mcp"] } } }

Tools exposed:

Tests

python -m pytest              # offline: normalization, param mapping, city resolution
OLXCARS_LIVE=1 python -m pytest tests/test_live.py   # hits OLX, incl. MCP round-trip

Notes and limits

• OLX's search API is undocumented and can change or vanish without notice. Every URL and parameter name lives in olxcars/client.py; if it breaks, that is the only file to change.

• OLX's own price_asc sort is not reliably sorted — treat sorting as a hint and sort client-side if exact ordering matters.

• One request per search; the location tree is cached for 7 days in ~/.cache/olxcars/locations.json.

• Scope: OLX India, cars (category 84). Other categories need their id from metadata.filters.