← Case Studies
Case Study — 02

AI Car Research
Agent

A Claude Code agent that automates used car research across Czech automotive portals — browsing listings, evaluating engine reliability, checking service history indicators, and generating structured buy / reconsider / avoid reports for every car it analyzes.

Type
Personal project
Stack
Claude Code · Playwright MCP
Portals
Sauto · AAA Auto · Auto ESA
Output
Markdown reports
The Problem

Buying a reliable used family car in the Czech Republic means browsing multiple portals, cross-referencing engine reliability data, evaluating service histories, and applying consistent criteria across dozens of listings. Done manually, this easily takes tens of hours — and the evaluation quality varies depending on how tired you are by listing number forty.

The harder problem is consistency. Every listing needs the same questions answered: Is this engine known to be reliable? Are there red flags in the service history? Does it meet family requirements? Without a structured protocol, important details get missed.

The Approach

A Claude Code agent with Playwright MCP as its browser — giving it the ability to navigate real automotive websites, extract listing details, and research reliability data from across the web. The agent follows a fixed evaluation protocol for every car, producing comparable reports regardless of the portal or model.

Two modes of operation: analyze a specific listing URL on demand (Task A), or run an active search across all three portals with defined filters and pull the best matches automatically (Task B).

How the Agent Works
01
Portal Navigation

The agent opens Sauto, AAA Auto, or Auto ESA with the defined filters — brand, year range, max mileage, price ceiling, fuel type, and transmission — then works through the results page by page.

02
Listing Extraction

For each candidate, the agent reads all available details: model, year, engine variant, mileage, price, seller type (dealer vs. private), service history notes, and any listed extras or damage history.

03
Engine Reliability Research

The agent evaluates the specific engine variant against known reliability data — distinguishing between problematic units (1.2 TSI chain issues, 1.4 TSI carbon buildup) and more dependable options (1.6 MPI, 1.5 TSI evo). Each engine gets a reliability note in the report.

04
Family Suitability Check

Every car is evaluated against a fixed checklist: rear ISOFIX points, boot capacity, Euro NCAP safety rating, and practical family usability — rear legroom, number of seats, and child-seat compatibility.

05
Report Generation

The agent writes a structured Markdown file for each car: basic info, service history assessment, reliability verdict, family checklist, red flags, and a final recommendation — Buy, Reconsider, or Avoid.

06
Summary & Ranking

After all cars are processed, the agent compiles a vysledky.md summary — a comparison table ranked by suitability score with top picks highlighted and pre-visit guidance for the buyer.

Output Format
  • Per-car report — one Markdown file per listing, named auto_[Model]_[Year]_[km].md
  • Basic info table — model, year, engine, mileage, price, seller type
  • Service history assessment — available records, gaps, known issues
  • Engine reliability note — variant-specific known issues and expected longevity
  • Family suitability checklist — ISOFIX, boot size, NCAP, legroom
  • Red flags — anything that warrants caution or further inspection
  • Verdict — Buy / Reconsider / Avoid with reasoning
  • Summary file — ranked comparison table across all analyzed cars
Stack
Claude Code Playwright MCP AI Agent Markdown
Key Outcomes
  • Consistent evaluation across all listings
  • Engine-specific reliability notes
  • Ranked shortlist with verdicts
  • Hours of manual research automated
Portals Covered
  • Sauto.cz
  • AAA Auto
  • Auto ESA