Skip to main content
Glama
README.md
# deptho-mcp

<!-- mcp-name: io.deptho/deptho-mcp -->

An MCP server that gives coding assistants (Claude Code, Claude Desktop, Cursor and others) Deptho's
backtest checks: replay a backtest's market orders against the trades that followed them, and write a
tearsheet that splits returns into fees, funding and slippage. It runs on your machine over stdio,
reads and writes local files, and sends nothing anywhere.

## Tools

| tool | does |
|---|---|
| `replay_orders` | orders CSV + trades file → fills CSV and a slippage summary (mean, median, p90, p99 bps) |
| `tearsheet` | bars CSV (`time`, `price`, `position`) → self-contained HTML report, its path and headline numbers |
| `pro_status` | whether Deptho Pro is installed and licensed here |
| `pro_sync` | Deptho Pro: download Binance USD-M days from the public archive into a local memory-mapped dataset |
| `pro_replay` | Deptho Pro: replay an orders CSV against that dataset |

**Orders CSV:** `time` (ISO 8601 or epoch), `side` (`+1`/`-1` or `buy`/`sell`), `qty` (base units, e.g. BTC).
**Trades file:** a Binance archive `aggTrades` file (`.zip` or `.csv`) or a Tardis.dev `trades` file
(`.csv` or `.csv.gz`) covering the orders' period.

Fills are modeled from trade prints: each order fills at the VWAP of the same-side trades after it until
its size is covered. Order book depth and queue position are not modeled.

## Install

```bash
pip install deptho-mcp
```

Use a virtual environment if you like; then point your assistant at that environment's `deptho-mcp`
(`which deptho-mcp` prints the full path). For the Pro tools, install Deptho Pro into the same
environment with your license key.

### Claude Code

```bash
claude mcp add deptho -- deptho-mcp
```

### Claude Desktop

Settings → Developer → Edit Config, then add to `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "deptho": { "command": "/full/path/to/deptho-mcp" }
  }
}
```

### Cursor

Add to `~/.cursor/mcp.json` (or `.cursor/mcp.json` in a project):

```json
{
  "mcpServers": {
    "deptho": { "command": "/full/path/to/deptho-mcp" }
  }
}
```

Then ask, for example: *"Replay orders.csv against BTCUSDT-aggTrades-2026-09-30.zip and tell me the
p90 slippage."*

## Privacy

Every tool works on local files. Nothing is sent to Deptho. The only download is `pro_sync`, which fetches
Binance's public archive (data.binance.vision) when you ask it to.

## Links

- Deptho: https://deptho.io/go/mcp
- Source: https://github.com/jjinaustin/deptho-mcp

License: Apache-2.0. Market data comes from Binance's public archive or from files you supply; Deptho is
not affiliated with Binance or Tardis.dev.

TDQS

A3.8/5.0

Scored across 5 tools

Disambiguation3/5

replay_orders and pro_replay have the same output and both replay an orders CSV, differing mainly by data source and Pro licensing. The distinction is explained, but an agent must reason about prerequisites to choose correctly. The remaining tools (tearsheet, pro_status, pro_sync) are clearly distinct.

Naming Consistency4/5

Most names use a readable verb_noun or prefixed form (replay_orders, pro_status, pro_sync, pro_replay), with a consistent pro_ prefix for licensed tools. The noun-only tearsheet is a minor deviation but remains clear as a domain term.

Tool Count5/5

Five tools is well-scoped for a backtest replay and reporting server. Each tool has a clear role: replay, reporting, and Pro dataset management.

Completeness4/5

The surface covers the core workflow: syncing trade data, replaying orders, and generating a tearsheet with headline metrics. Minor gaps exist around inspecting intermediate results or non-Pro dataset management, but agents can work around them.

Maintenance

ActivityMaintained
ResponsivenessSyncing