Connect your own agent to the TTC API for real-time market data and begin perp trading across CEX | DEX exchanges in a standardized way.
The ttc CLI ships with all skills pre-bundled under .claude/skills/. Claude Code auto-discovers them — no loader, no prompt injection, no schema glue. Install the binary, open the repo, and the skills are live.
# install the CLI cargo install ttc-cli # open the project in Claude Code claude # skills listed below are available immediately: # skill-trading, skill-dca, skill-twap, skill-momentum, # skill-market-maker, skill-loop-trading, skill-shark, # skill-signal-patrol, skill-portfolio-manager, # skill-market-overview, skill-onboarding, skill-claude-diag
Why this is the best path: Claude Code is a flat-rate terminal agent — you are not paying per-LLM-call through an API gateway, and CLI tools use ~25× fewer tokens than MCP for the same task.
For any MCP-capable host (Claude Desktop, Cursor, Windsurf, custom agents): point at the TTC MCP endpoint and pass your session headers — the server forwards ttc-auth-token and ttc-public-key through to the exchanges API so paid tools resolve without an x402 payment. Skills become callable tools in the agent's toolbox.
Remote (Streamable HTTP)
{
"mcpServers": {
"ttc": {
"type": "http",
"url": "https://ttc.box/api/v1/mcp",
"headers": {
"ttc-auth-token": "${TTC_AUTH_TOKEN}",
"ttc-public-key": "${TTC_PUBLIC_KEY}"
}
}
}
}Export TTC_AUTH_TOKEN and TTC_PUBLIC_KEY in your shell (never hard-code them).
Local (stdio)
{
"mcpServers": {
"ttc": {
"command": "ttc",
"args": ["mcp"]
}
}
}Works everywhere MCP works — but expect higher token usage and a stateful connection per session.
Building your own agent loop? Read the SKILL.md files from disk, inject their frontmatter into the system prompt, and invoke the ttc binary as a subprocess when the model calls a skill. That is the entire integration.