Choose a Setup

Choose between Hosted MCP, Local MCP, CLI, Remote MCP, and Agentic Skills based on security and workflow requirements.
View as Markdown

Choose a setup based on two questions:

  1. Which AI client do you want to use?
  2. Do you need public data only, private account access, or trading workflows?

If you are not sure, start with Hosted MCP for public data or Local MCP for private account workflows.

Pick by use case

Quick decision table

GoalRecommended setupWhy
Ask an AI client for public market dataHosted MCPNo local install and no API key required.
Let Claude Desktop, Cursor, or ChatGPT Desktop call 6MM toolsLocal MCPAI calls tools through a local server controlled by the user.
Let Codex, Claude Code, Gemini CLI, or scripts call 6MMCLICommand output can be JSON and easy for agents to parse.
Provide a company-managed MCP endpointRemote MCPThe institution controls hosting, credentials, access, and logs.
Standardize repeated analysis or risk workflowsAgentic SkillsTemplates define tool usage, guardrails, and output format.

Pick by AI client

AI clientHosted MCPLocal MCPCLINotes
Claude DesktopPossibleRecommendedPossibleUse Local MCP when account context is needed.
Claude.ai WebRecommended when supportedNot applicableNot applicableBest for public market data only.
ChatGPT DesktopPossibleRecommendedPossibleLocal MCP is preferred for private account tools.
CursorPossibleRecommendedPossibleLocal MCP works well with coding workflows.
Codex CLIPossiblePossibleRecommendedCLI is usually the most direct path.
Claude Code CLIPossiblePossibleRecommendedCLI or Local MCP depending on workflow.
VS Code / CopilotPossibleRecommendedPossibleDepends on MCP and terminal support.
Gemini CLIPossiblePossibleRecommendedCLI is the simplest first setup.
WindsurfPossibleRecommendedPossibleLocal MCP is preferred.

Pick by setup path

SetupInstallationAPI key locationBest initial scopeAccount and trading scope
Hosted MCPNoneNo user API key for public dataPublic market dataNot recommended for first release
Local MCPUser installs packageUser machinePublic data and account-readRecommended path for future trading
CLIUser installs packageUser machinePublic data and scriptsUseful for preview and automation
Remote MCPUser deploys serverUser cloud or institution cloudPublic data or account-readPossible when the user controls deployment
SkillsDocs or installable templatesDepends on underlying toolsAnalysis workflowsMust follow tool permissions

Security rule of thumb

Use the lowest-risk setup that supports the task:

  • Public market data: Hosted MCP is enough.
  • Private account data: Local MCP or user-managed Remote MCP.
  • Trading preview: Local MCP or CLI, preferably testnet first.
  • Real trading: only after explicit permission, preview, confirmation, limits, and audit logs.

Next step

After choosing a path, open the matching guide and complete one verification prompt before adding sensitive credentials.