io.github.mohammad-abbasi-sh/paribu-mcp
# paribu-mcp πΉπ·
mcp-name: io.github.mohammad-abbasi-sh/paribu-mcp
Unofficial **read-only** MCP server for **Paribu** (paribu.com), TΓΌrkiye's largest crypto exchange β live 24h tickers for 266 markets (BTC, ETH and 190+ coins vs TL and USDT), order books with spread and mid price, recent trades, the full 467-market catalogue, and currency metadata with networks and withdrawal fees.
> Part of the "Turkey MCP factory". Not affiliated with Paribu. Data comes from Paribu's official public API (`api.paribu.com`, docs at `docs.paribu.com/api`) β no login, no API key, read-only.
## For AI agents
- **Market ids** are lowercase `base_quote`, e.g. `btc_tl` β the Turkish lira quote is spelled `tl`, not `try`.
- **Ticker**: `paribu_get_ticker` with a market returns one 24h snapshot; without a market it returns a ranked list (default sort: pair volume). 266 markets have live tickers; the config lists 467 pairs (some are OTC-only or not yet trading).
- **Order book**: `paribu_get_orderbook` returns best bid/ask, spread, mid price and the levels. An unknown market answers with an error β Paribu itself returns HTTP 200 with a null body for it.
- **Trades**: `paribu_get_recent_trades` β limit 1β20 (the API maximum), newest first.
- **Catalogue**: `paribu_list_markets` lists all 467 pairs with price/amount precision and steps; `paribu_get_currency` returns a currency's networks, deposit/withdrawal limits and per-network withdrawal fees.
- **Movers**: `paribu_market_movers` ranks the top gainers/losers by 24h percentage.
- Every response carries `fetched_at` (UTC) and `source: api.paribu.com`.
## Install
```bash
uvx paribu-mcp
```
Or add to your MCP client config (e.g. Claude Desktop):
```json
{
"mcpServers": {
"paribu": {
"command": "uvx",
"args": ["paribu-mcp"]
}
}
}
```
## Tools
| Tool | What it does |
| --- | --- |
| `paribu_get_ticker` | 24h ticker for one market, or a ranked list of all 266 markets (filter by `tl`/`usdt`, sort by volume/change/%) |
| `paribu_get_orderbook` | Live order book: best bid/ask, spread, mid price, levels (1β100 per side) |
| `paribu_get_recent_trades` | Most recent public trades, newest first (1β20) |
| `paribu_list_markets` | Full 467-pair catalogue with precision, steps, labels, OTC flag |
| `paribu_get_currency` | Currency metadata: networks, arrival times, explorers, withdrawal fees, limits |
| `paribu_market_movers` | Top gainers / losers over the last 24h |
| `paribu_health` | Is the Paribu public API reachable right now? |
## Example
```json
{ "market": "btc_tl", "last": 3252946.0, "low": 3246001.0, "high": 3344137.0,
"change": -84957.0, "percentage": -2.54, "volume": 32.147208,
"pair_volume": 106109641.0 }
```
## Caveats
- Prices are live market data, not investment advice β verify on paribu.com before acting on them.
- An unknown market id on the ticker endpoint returns HTTP 500 with code `5001`; on the order book it returns HTTP 200 with a null body.
- The config endpoint is polled at most once per call; Paribu recommends polling it no more than every 60 seconds.
- Transport uses Chrome TLS impersonation with a plain-requests fallback; requests are spaced ~0.5s apart.
## Development
```bash
python -m venv .venv && .venv/bin/pip install -e ".[dev]" # or: pip install -e . pytest
.venv/bin/python -m pytest tests/test_markets.py -v # offline, recorded fixtures
.venv/bin/python -m pytest tests/test_live.py -v # live, against api.paribu.com
.venv/bin/python tests/smoke_live.py # daily CI probe
```
Built by [Mohammad Abbasi](https://github.com/mohammad-abbasi-sh) β [Almanibama](https://almanibama.ir) Β· [mabbasi.ir](https://mabbasi.ir)
TDQS
Scored across 7 tools
Each tool targets a distinct data shape: snapshot ticker, live order book, recent trades, market catalogue, currency metadata, movers, health. The only real overlap is between paribu_get_ticker (which can sort by 'percentage'/'change' across all markets) and paribu_market_movers (ranked gainers/losers), which an agent could plausibly pick the wrong one for.
A clear 'paribu_' namespace plus verb_noun convention (get_ticker, get_orderbook, get_recent_trades, list_markets, get_currency) is followed for most tools. Two outliers β paribu_market_movers and paribu_health β drop the verb, a minor deviation that stays readable.
Seven tools is well within the ideal range and each covers a distinct slice of a read-only exchange data API. Nothing feels padded or redundant.
The surface covers the core public market-data lifecycle: discovery (list_markets, get_currency), pricing (ticker, orderbook), activity (recent_trades, movers) and availability (health). The notable gap is historical/OHLC candlestick data, which agents would expect from an exchange data server but must work around.