Ozon product-research MCP in Rust
Ozon MCP
A local Rust MCP server for careful Ozon product research: search, cards, reviews and photos through an anonymous session, no account login and no cart changes
Low risk
We rate an entry low when it mostly gives the agent instructions and reference material.
Why this level
- Runs an anonymous session and never changes the cart, orders, account or region
Install
Manual install
cargo build --release --locked && cargo install agent-browser --version 0.36.0 --locked && agent-browser installBuilding the server and installing the required browser backend version.
This is third-party code. Review the repository files before installing.
What it does
The server runs an embedded agent-browser and keeps one observed context and a persistent profile shared by several MCP clients through a local broker. Tools cover context and region observation, search with refinement, batch product reads, reviews with aggregates, verified images from saved links, and a local research log with idempotent agent notes. Unknown and incomplete data comes back as unknown, partial, or a typed error rather than a guess, and the Ozon Card price is kept separate from the regular one. The server never changes the cart, orders, account or region, and explicitly calls itself development in progress, validated locally across 30 calls.
Who it is for. For people who want an agent to do careful, anonymous Ozon product research (comparison, reviews, photos) with no risk of ordering or changing anything in a real account.
Good fit when
- You need Ozon product research against requirements, budget and reviews without an account login
- An anonymous session defaulting to Moscow works for you
- You are on macOS or Linux and can install Rust and agent-browser 0.36.0
Not a fit when
- You need other sellers' offers for the same item: not supported by the observable source yet
- You need Windows: the server does not support that platform
- You need account, cart or order access: the server deliberately never does this
Example request
Find wireless headphones on Ozon under 5000 rubles with good noise-cancellation reviewsLimitations
The project status is explicitly development, not a stable release, and the author openly states it is based on Pir0manT/ozon-mcp-server and the original eduard256/ozon-mcp-server, with git history and attribution preserved. It needs Rust 1.95 and exactly agent-browser 0.36.0. Only macOS and Linux are supported. The browser profile and data directory must live outside the repository and be owned by the current user.
How to disable. Remove the ozon block from your MCP client's mcpServers config and stop the server process.
MCP
- Transport
- stdio
- Authentication
- not required
| Environment variables | |
|---|---|
| OZON_DATA_DIR required | Directory for the private IPC and research log, 0700 permissions |
| OZON_HEADLESS | Whether the browser runs headless, false is needed for manual login or changing region |
Security check
- Runs an anonymous session and never changes the cart, orders, account or region
README in short
The README describes the tools by group (context, search, cards, reviews, images, research log), configuration via OZON_DATA_DIR, OZON_USER_DATA_DIR, OZON_HEADLESS and other environment variables, the manual-login behavior for changing region, development commands (fmt, test, clippy, page checks), and explicitly names the project's origin from Pir0manT and eduard256.
FAQ
Can the server place an order?
No, the server only reads Ozon and never changes the cart, orders, account or region.
Is this independent work?
No, the author explicitly states the project is based on Pir0manT/ozon-mcp-server and the original eduard256/ozon-mcp-server, with git history preserved.
Related
A self-hosted knowledge base with block-level references and a built-in MCP server for connecting AI agents to your notes
A CLI for every Google Workspace API with JSON output and agent skills: Drive, Gmail, Calendar, Sheets and more
Local search over Markdown notes, docs and meeting transcripts: keywords, semantic search and reranking, with an MCP server
A task manager for AI-driven development: breaks a PRD into dependent tasks and guides the agent through them via MCP or CLI