The Model Context Protocol (MCP) is an open standard that lets an AI assistant use external tools. The platform ships an MCP server so your own agent — Claude Desktop, Cursor, or any MCP-compatible client — can drive it: build a strategy, run a backtest, spin up a Squad, all in plain language, through 70+ tools grouped by area (strategies, indicators, signals, bias, risk, backtest, grids, versioning, reference, and agents).
This is the counterpart to the gateway: the gateway is your own hardware executing trades; MCP is your own agent controlling the platform.
How it fits together#
The MCP server uses stdio transport, the universal standard every major client supports. It authenticates to the platform with your personal token, set up once through a one-click browser flow (see installation).
What you can do#
- Create and configure strategies (CRUD + versioning)
- Add/update/remove indicators with parameters
- Define signal rules and bias generators
- Configure stop-loss and take-profit
- Run backtests and review metrics
- Create grid search templates + execute them
- Manage strategy versions and restore snapshots
- Share strategies to community
- Browse, clone, and run hedge-fund Squad templates
- Inspect Squad deliberation messages + track records
- Manage LLM connections, model routes, and daily spend caps
- Pull news articles, macro events, and Squad decisions
Supported clients#
- Claude Desktop
- Cursor
- Windsurf (Cascade)
- VS Code (Copilot)
- Cline (VS Code extension)
- Continue.dev
- Hermes (Nous Research)
- OpenClaw
- Any MCP-compatible client
The tools#
The 70+ tools cover the whole platform surface — core strategy building (strategies, indicators, signals, bias, risk, backtest, grids, versioning), the multi-agent layer (Squad templates, deliberations, model routes, LLM connections, spend caps, news, macro), and community sharing. The full, current catalogue with every argument is the tools reference — it is generated from the server itself, so it never drifts from what the tools actually accept.