With the latest advances in AI, QuantConnect has never been easier to use. You can implement almost any strategy you could conceive on a robust, battle-tested platform. 

Today we're officially announcing our remote MCP, an easy way to connect your local Claude Code, Codex, or another compatible AI client directly to our remote MCP server.

LLMs need the right tools and context to do useful work. Our MCP equips your preferred model with access to QuantConnect’s data, project files, research environment, and compute. Dataset examples, project templates, and compilation feedback help the model be productive on QuantConnect and maintain its focus on the research problem. 

The remote connection lets you bring models such as Fable 5.1, Opus 5.5, and Astra through your preferred MCP-compatible client, with QuantConnect providing the tools and infrastructure for quantitative research. The MCP provides more than 45 tools covering the quant research workflow:

  • Agents can create projects, read and edit Python or C# files, and compile changes to check for errors.
  • They can explore the Data Library, inspect dataset examples, and create and execute Jupyter notebooks in the Research Environment.
  • They can run backtests, inspect results, orders, and logs, and test parameter combinations through optimizations.
  • They can deploy strategies to paper trading and inspect existing live algorithms, including their holdings, performance, and logs.
  • They can write files to the Object Store, convert stored PDFs into Markdown for analysis, and send email, SMS, or Telegram notifications.

You can connect agents such as Dots, Muse, or Duet via the remote MCP to conduct research on your behalf. For parallel research, assign separate projects to different agents and expand your agent-node capacity to support more concurrent tasks. Each agent can investigate its own hypothesis and produce notebooks and backtests for review.

To connect with the Remote MCP, add https://www.quantconnect.com/api/v2/mcp to your client and authorize your QuantConnect organization. Setup guides are available for Claude Code, Copilot, Cursor, and Codex.

Remote MCP access requires a paid A1-1, A4-8, or A16-32 agent node in your organization. See the resource documentation to add a node.

We also serve a cloud hosted harness (think - "Claude Code for Quant"). Our bespoke harness is customized for quant development and deeply integrated with QuantConnect. The cloud hosted Agents support coordinated work within a project, with Agent Teams that communicate via messages, share findings among specialists, and work toward a shared objective. The Conductor coordinates this process, bringing in specialists such as the Research Agent and Backtest Agent as needed. You can customize their instructions, configure tasks, and bring your own model-provider API key. The cloud hosted harness maintains the conversation and agent state across long research runs, including while backtests and optimizations complete.

We think you'll be blown away at the combination of frontier model intelligence, and our rich MCP. If you've ever wanted to give quant a try, there's never been a better time. 

Happy Trading!

QuantConnect Team