Backtest360

What Backtest360 MCP server does

Backtest360 MCP server connects an MCP client to the Backtest360 engine API. It gives an agent structured tools for exploring supported indicators, operators, execution modes, stop types, sizing methods, bar frequencies, and metric sections. Agents can retrieve strategy templates and schemas, validate strategy documents without executing them, run historical backtests, compare multiple strategies on the same data, evaluate the latest bar, and calculate statistics from an external returns series.

The server also exposes asset and market-data operations. These include ticker search, data-range estimates, ticker metadata, latest quotes, historical OHLCV data, and macroeconomic series catalogs and observations. Quote and historical price retrieval require the relevant paid capability. The API key’s plan determines available permissions, limits, and data access.

How it works

The server is a thin adapter over the engine’s HTTP API. It does not contain an AI model and does not independently calculate backtest results. The connected client decides how to interpret tool responses and can use the supplied workflow prompts to scaffold common sequences such as building and validating a strategy or reviewing robustness.

Backtest results can be shaped to reduce context size. A request can return a summary, statistics, or fuller output with selected trades, equity, return, and signal-diagnostic blocks. Long series may be downsampled, and oversized results are reduced further with an explicit truncation marker rather than being silently cut. Validation failures and request-level engine rejections are returned as structured results when changing the request may fix the problem.

Setup and configuration

Backtest360 MCP server requires Python 3.10 or later for local installation and a Backtest360 API key. Install the package with pip install backtest360-mcp, then run the backtest360-mcp command for a stdio connection. Alternatively, configure an MCP client to use https://mcp.backtest360.com/mcp as a streamable-HTTP endpoint and send the key in the X-API-Key header.

Local configuration uses these environment variables:

• BACKTEST360_API_KEY: required engine key.
• BACKTEST360_ENGINE_URL: optional engine base URL; defaults to https://api.backtest360.com.
• BACKTEST360_MCP_TIMEOUT: optional per-request timeout in seconds; defaults to 300.
• BACKTEST360_MCP_MAX_OUTPUT_BYTES: optional maximum size for one tool result; defaults to 100000.

Tools and capabilities

The available tools cover four main workflows:

• Inspect engine health, API details, key permissions, usage, and capability flags.
• Discover indicators, templates, schemas, and reference catalogs before constructing a strategy.
• Validate, backtest, compare, and evaluate strategies, then compute metrics from supplied returns.
• Discover symbols and retrieve eligible market or macroeconomic data.

Static catalogs and the strategy schema are also available as MCP resources for clients that support resource attachment. Two prompts describe tool order for strategy construction and robustness review, but the connected AI remains responsible for interpretation.

Limitations and notes

The Backtest360 MCP server depends on the configured engine key and its plan. Server-side historical price access is a paid capability, while the free tier can run backtests on data uploaded to the engine. Rate limits, permissions, engine availability, and compute timeouts can affect requests; error responses include guidance such as retry timing, missing capabilities, or scope reduction where applicable.

The adapter does not turn raw results into investment conclusions. Signal diagnostics are capped lists of condition fire dates, not raw per-bar boolean arrays. Full results can be large, so response detail, selected output blocks, and series limits should be chosen according to the client’s context capacity.