candlestick-chart-mcp
Renders OHLCV data as candlestick chart PNGs and returns inline images. It is a pure renderer with no network access, allowing models to view charts generated from provided data.
README
candlestick-chart-mcp
An MCP server that renders OHLCV data as a candlestick chart PNG and returns it as inline image content, so the model can actually look at the chart it asked for.
It is a pure renderer: no network access, no API keys, no market-data provider. You pass the bars in; it draws them.
Install
npm install
npm run build
Verify it end to end (spawns the server over stdio and calls the tool for real):
npm run check
Render sample charts to out/ to eyeball the output:
npm run smoke
Register with a client
Claude Code:
claude mcp add candlestick-chart -- node /absolute/path/to/candlestick-chart-mcp/dist/index.js
Or in a config file (.mcp.json, claude_desktop_config.json, …):
{
"mcpServers": {
"candlestick-chart": {
"command": "node",
"args": ["/absolute/path/to/candlestick-chart-mcp/dist/index.js"]
}
}
}
The tool
render_candlestick_chart
Returns two content blocks: a short text summary (bar count, date range, OHLC, percent change, plus any notes about inputs that were adjusted) and the PNG.
Bars. candles accepts either form per bar, and mixing them is fine:
{ "time": "2026-01-05", "open": 100, "high": 103.2, "low": 99.4, "close": 102.8, "volume": 1200000 }
[1767571200, 100, 103.2, 99.4, 102.8, 1200000]
The tuple form is [time, open, high, low, close, volume?] and costs far fewer
tokens — prefer it for anything longer than a handful of bars. time takes epoch
seconds, epoch milliseconds, or any string Date.parse understands. Bars are
sorted by time; a high/low that contradicts the body is clamped rather than
rejected.
Everything else is optional:
| Option | Default | Notes |
|---|---|---|
title, subtitle |
— | Subtitle defaults to date range · bar count · change |
chartType |
candlestick |
or heikin-ashi, ohlc-bars |
width, height, scale |
1400, 800, 1 |
scale is the device-pixel ratio |
theme |
dark |
or light |
colorScheme |
diverging |
or classic, mono — see below |
bodyStyle |
per scheme | filled or hollow-up |
showVolume |
true |
Needs a volume column; drawn in its own panel |
showGrid, showLegend, showLastPrice |
true |
|
yScale |
linear |
log for multi-year or high-growth ranges |
timeZone |
UTC |
Any IANA zone; applies to every label |
sma, ema |
[] |
Arrays of periods, e.g. [20, 50] |
bollinger |
— | { "period": 20, "stdDev": 2 } |
vwap |
false |
Session-anchored; needs volume on every bar |
priceLines |
[] |
{ price, label?, color?, style? } — support, resistance, stops |
markers |
[] |
{ index? or time?, side: buy/sell/neutral, label? } |
Example call:
{
"candles": [[1767571200, 100, 103.2, 99.4, 102.8, 1200000], "…"],
"title": "BTCUSD · 1D",
"sma": [20, 50],
"bollinger": { "period": 20, "stdDev": 2 },
"priceLines": [{ "price": 61941.5, "label": "resistance" }],
"markers": [{ "time": "2026-02-14", "side": "buy", "label": "entry" }],
"yScale": "log",
"timeZone": "Europe/Riga"
}
Inputs that cannot be honoured do not fail the call — an SMA period longer than the series, a missing volume column, a log scale with non-positive prices, an unknown time zone are each dropped or downgraded and reported in the text block.
Colour
Up/down is a polarity encoding, so the default diverging scheme uses the
blue ↔ red diverging pair. That pair clears the all-pairs colourblind gate in
both themes (CVD ΔE 21.6 light / 19.2 dark against a ≥ 8 target). The traditional
green/red pair fails it at ΔE 4.1 for deuteranopia, so classic ships with
hollow up-bodies: shape carries the direction and hue only reinforces it.
mono drops hue entirely — ink, hollow versus filled — for print and
forced-colours contexts.
With the up/down pair already on screen there is no room left for a palette of overlay hues: every candidate measured below the normal-vision floor against one of the poles. So all overlays share one validated accent per scheme, and their identity comes from dash pattern + legend + a direct label at the end of each line. Four overlays are distinguishable; a fifth reuses a dash pattern and the response says so.
Volume bars inherit the bar's direction at half strength so the panel stays recessive, and it is a separate panel — never a second y-axis on the price plot.
Layout notes
- Bars sit on an ordinal x-axis (one slot per bar), so weekends and market holidays leave no gaps. Time labels appear where the label actually changes; intraday charts substitute the date wherever the day rolls over.
- Below ~3.5 px per bar, bodies are dropped and each bar becomes a single high-low hairline — a 1200-bar chart stays readable.
- Bodies cap at 26 px wide so a seven-bar chart still looks like candles.
- The PNG is capped near 4.5 MB; a request that would exceed it is re-rendered at
scale: 1and, failing that, returns an error asking for smaller dimensions.
Layout of the source
| File | Role |
|---|---|
src/index.ts |
MCP server, tool registration, marker/time-zone resolution |
src/schema.ts |
Zod input schema — the tool's contract |
src/render.ts |
All drawing: layout, axes, candles, overlays, annotations |
src/theme.ts |
Palettes and the validation results behind them |
src/indicators.ts |
SMA, EMA, Bollinger, VWAP, Heikin-Ashi |
src/candles.ts |
Input normalisation |
src/format.ts |
Tick stepping, number and date formatting |
Adding an indicator panel (RSI, MACD) means adding a panel band to the layout
block in renderChart and a series function in indicators.ts.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.