How do I install the KLinePic stdio MCP server directly from GitHub?

[sher1096]

I want an AI agent to turn broker or exchange fills into annotated post-trade candlestick-chart PNGs, but I do not want to clone a repository or depend on a published npm package.

KLinePic provides a first-party stdio MCP adapter with three tools:

• klinepic_get_capabilities checks key permissions and remaining quota.
• klinepic_preflight_review_chart validates a request without consuming render quota.
• klinepic_create_review_chart returns the rendered PNG as native MCP image content.

The source is public under MIT: https://github.com/sher1096/klinepic-agent-api-examples

The remote GitHub installation path has been tested with the official MCP SDK: initialization succeeds and tools/list returns all three tools. The production preflight path has also been tested with chargeable=false.

What is the minimal configuration for an MCP client that supports stdio and environment variables? Is there a Windows-specific command form, and how can I verify the setup without consuming chart quota?

Requirements and boundaries:

• Node.js 20 or newer.
• A self-service Agent API key from https://klinepic.com/account/agent?utm_source=github&utm_medium=referral&utm_campaign=geo202607
• KLinePic is for post-trade review and visualization. It does not provide trading signals, price predictions, or order execution.

[sher1096]

Use the GitHub repository itself as the npx package source.

macOS / Linux and clients that resolve npx directly

{
  "mcpServers": {
    "klinepic": {
      "command": "npx",
      "args": ["-y", "github:sher1096/klinepic-agent-api-examples"],
      "env": {
        "KLINE_AGENT_API_KEY": "kline_agent_xxx"
      }
    }
  }
}

Windows fallback

Some Windows MCP clients launch commands without a shell and cannot resolve npx. In that case, route it through cmd.exe:

{
  "mcpServers": {
    "klinepic": {
      "command": "cmd",
      "args": [
        "/c",
        "npx",
        "-y",
        "github:sher1096/klinepic-agent-api-examples"
      ],
      "env": {
        "KLINE_AGENT_API_KEY": "kline_agent_xxx"
      }
    }
  }
}

Restart the MCP client after saving the configuration. Then verify the setup in this order:

1. Call klinepic_get_capabilities. This confirms that the key is recognized and shows permissions and quota.
2. Call klinepic_preflight_review_chart with { "request": <Agent API request body> }. Use the repository's sample request rather than rebuilding the payload from memory.
3. Confirm that the response is valid and reports the preflight as non-chargeable.
4. Only then call klinepic_create_review_chart; this is the step that renders an image and consumes chart quota.

Useful checks:

node --version
npx -y github:sher1096/klinepic-agent-api-examples

Node.js must be version 20 or newer. The server can start without a key so a registry can inspect the tool schemas, but live capability, preflight, and render calls require KLINE_AGENT_API_KEY.

If you prefer an isolated process, clone the repository and run the included Dockerfile:

docker build -t klinepic-mcp .
docker run --rm -i -e KLINE_AGENT_API_KEY=kline_agent_xxx klinepic-mcp

For request fields, quota rules, and self-service key creation, use the Agent API guide. For source=csv, include klineCsv with enough rows to cover the requested display window.