Helix MCP Server

What you get

Defensive chart generation across the full graph pipeline, even with arbitrary user data.
Plugin-aware responses so the agent can pick chart styles, data platforms, exports, and AI controllers.
PNG + JSON in one response with pipeline metadata for auditing and retries.
Discovery endpoints (get_server_info, list_graph_capabilities) so agents self-configure.

Set up in Claude Desktop

Install dependencies (includes mcp>=1.17.0): python -m pip install -r requirements.txt
Ensure your Helix account has an active subscription (the server calls /api/mcp/validate on start).
Open Claude Desktop > Settings > Developer > MCP and add a server:

{
  "command": "/absolute/path/to/.venv-14/bin/python",
  "args": ["-m", "mcp"],
  "transport": "stdio",
  "env": {
    "ASKFELIX_API_SECRET": "your-secret",
    "ASKFELIX_API_BASE_URL": "https://helixapi.app.nz",
    "ASKFELIX_MCP_DOC_URL": "https://helixapi.app.nz/mcp"
  }
}

Restart Claude, then run get_server_info to confirm the MCP server is reachable and the secret maps to your account.

Set up in Codex or other MCP CLIs

Add an entry to ~/.config/codex/mcp.toml (adapt the command path to your interpreter):

[servers.helix]
command = "/absolute/path/to/.venv-14/bin/python"
args = ["-m", "mcp", "--transport", "stdio"]
env.ASKFELIX_API_SECRET = "your-secret"
env.ASKFELIX_API_BASE_URL = "https://helixapi.app.nz"
env.ASKFELIX_MCP_DOC_URL = "https://helixapi.app.nz/mcp"

Run codex mcp test helix to verify connectivity before issuing chart prompts.

Make your first graph

Give the agent a small CSV or table. It can be attached, pasted, or referenced by URL the MCP server can reach.
Ask explicitly for a chart and an image: "Use the helix MCP server to plot revenue by month as a PNG and include the response JSON."
The agent will call generate_graph, select plugins (chart type, export target, data platform), and return the PNG plus JSON in one response.

# Direct HTTP equivalent (useful for debugging)
curl -X POST https://helixapi.app.nz/api/mcp/generate_graph \
  -H "Content-Type: application/json" \
  -d '{
        "secret": "your-secret",
        "prompt": "Bar chart of revenue by month; return a PNG image.",
        "output": "png",
        "include_response_json": true
      }'

If your agent prefers JSON, set "output": "json" and let it render locally; PNG is fastest for quick previews.

Tools and controls

Tool	Purpose	Agent controls
`generate_graph`	Send natural language chart requests and receive PNG or JSON.	chart plugin, export target (png/json/svg), AI controller, data platform, pipeline overrides.
`list_graph_capabilities`	Enumerate plugins and export targets so the agent can plan before calling `generate_graph`.	filters for plugin families (chart/export/dataplatform).
`get_server_info`	Return base URL, default output, timeout, and documentation link for self-configuration.	none (read-only discovery).

All inputs are Pydantic-validated to keep arbitrarily large or messy user data safe while giving the agent fine-grained control.

Troubleshooting