Does MarketLens have an MCP server?

Yes. Install the optional extra with pip install 'marketlens[mcp]' and run the marketlens-mcp console script. It serves the SDK over stdio with your API key, so any MCP client (Claude Code, Claude Desktop, Cursor) can research Polymarket markets, pull order books and surfaces, and run backtests in natural language.

Integration

MCP Server

Expose the SDK to any MCP client so an agent can author and run backtests, research markets, and pull order book data and surfaces in natural language. Optional extra, no effect on the base install.

Setup

Install the optional extra and add the server to your MCP client config. It runs locally over stdio and authenticates with your API key, the same one the SDK and REST API use.

Works with Claude Code, Claude Desktop, Cursor, and any other MCP client. Compose it alongside other servers (arxiv, web search, filesystem) so agents can read the literature, pull data, and backtest in a single session.

bash

$ pip install 'marketlens[mcp]'
# or python -m marketlens.mcp

jsonmcp client config

{
  "mcpServers": {
    "marketlens": {
      "command": "marketlens-mcp",
      "env": { "MARKETLENS_API_KEY": "mk_..." }
    }
  }
}

Tools

Data tools wrap the typed resources; list tools return compact rows capped at 200 per call.

Data

search_markets, get_market

Find and inspect markets.

search_events, search_series

Browse events and recurring series.

get_orderbook

Point-in-time L2 book with spread, microprice, and imbalance.

get_orderbook_metrics

Time-bucketed book metrics over a window.

get_trades, get_candles

Executed trades and OHLCV.

get_reference_candles

Binance spot for the underlying asset.

get_signals, get_surface

Implied probability surfaces.

Backtesting

strategy_reference

The Strategy / StrategyContext API, so the agent writes correct code.

run_backtest

Run agent-authored strategy code through the engine. Returns metrics and a saved result path.

compare_backtests

Score several strategies over one target and window side by side.

open_backtest

Inspect a saved run's metrics, per-series PnL, and settlement ledger without re-running.

Environment

MARKETLENS_API_KEYrequired

Your API key. Authenticates every tool call.

MARKETLENS_MCP_DISABLE_BACKTEST

Set to 1 to disable local strategy execution.

agent session

user
  Backtest a mean-reversion entry on
  btc-up-or-down-5m: go long YES when the
  midpoint dislocates below 0.35, size 200
  shares, then liquidate at resolution. Replay
  2026-04-15 12:00 to 18:00 UTC, $10,000 capital.

agent
  strategy_reference()
  run_backtest(
      "btc-up-or-down-5m",
      after="2026-04-15T12:00:00Z",
      before="2026-04-15T18:00:00Z",
      initial_cash=10_000,
      data_dir="data",
  )

  total_return  +4.1%
  win_rate      58%
  sharpe        1.9
  saved → runs/btc-dip-buyer