LuxAlgo MCP

 

Library · Brokers · Edge Stats · Market Trackers · Challenge Simulator · Prop Firms · Vela for Developers · npm · Endpoint

LuxAlgo MCP is a LuxAlgo open-source project. Overview and setup: luxalgo.com/mcp.

It puts the LuxAlgo ecosystem behind a single MCP server: an encyclopedia of trading and technical analysis, your LuxAlgo trade journal, read-only access to your own brokerage accounts, hosted session statistics with a sample size on every number, the public record of US markets (congressional trades, insider filings, lobbying, contracts, patents and more, with a primary-source link on every row), a Monte Carlo challenge simulator, and a live prop-firm directory. No API key for anything hosted: the public tools are free and read-only, the Trade Journal and account tools use your LuxAlgo sign-in, and the local broker tools use your own keys and never send them anywhere.

claude mcp add --transport http luxalgo https://mcp.luxalgo.com/mcp

What's inside

Area	What you get
Library	The encyclopedia of trading and technical analysis: hundreds of concept pages with formulas, the full indicator catalog with families and tags, and Pine Script® sources where publicly served.
Trade Journal (sign-in)	Your own journal in the LuxAlgo app: the dashboard (metrics, Edge Score, equity curve), the P&L calendar, breakdowns by weekday / hold time / symbol / tag and more, every trade with its fills and annotations, day notes — and the writes that keep it alive: log fills by hand, annotate trades (tags, mistakes, rating, stop and target, review), write notes. Always as you, on your data; the app owns the rules.
Brokers (local only)	Read-only access to your own accounts across 22 brokers and exchanges via broker-sdk: balances, positions, trade history, FIFO performance stats. Keys live in your MCP client config as env vars and never leave your machine. The hosted endpoint does not carry these tools, on purpose.
Edge Stats	Hosted session statistics from the open-source edge-stats engine: how often a setup actually worked (gap fills, opening-range breakouts, day-of-week effects, event days) with the sample size and a Wilson 95% confidence interval on every number. A nightly build runs the real engine over free market data and publishes only derived statistics; these tools serve them verbatim.
Market Trackers	The public record of US markets from primary sources only: congressional trades, insider (Forms 3/4/5) transactions, 13F holdings, federal contracts and grants, lobbying filings, FINRA short-sale volume, granted patents, clinical trials, FDA drug events, CFTC positioning, federal bills, FEC campaign finance, hearing transcripts, Federal Reserve communications, committee assignments, Wikipedia pageviews. Read straight from the pipeline's CC0 dumps — live tree plus deep-history archives — with provenance.sourceUrl on every row. Data only: no signals, scores, or predictions.
Challenge Simulator	The open-source prop-firm-sim Monte Carlo engine, running locally inside the server. Your stats, or your real R-multiple trade series, through a firm's exact ruleset: pass probability with confidence intervals, expected attempts and cost, EV over the funded horizon, optimal-risk sweeps, cross-challenge comparison. Deterministic under seed, every assumption disclosed.
Prop Firm Directory	The live data the simulator draws from: firms, funded-account challenges with their full rulebooks (account sizes, fees, steps, profit splits, drawdown modes, trading restrictions), and current offers.
Charts (your browser)	Not a tool: the chart you draw with what the tools return. The open-source core of Vela by LuxAlgo runs the Pine Script® that library_get_source_code hands back and paints the fills that broker_trades lists, in a browser tab, on your machine. How the loop works.

Install

The hosted server is one URL:

https://mcp.luxalgo.com/mcp

Claude (web, desktop, mobile)

Customize → Connectors → Add custom connector, URL https://claude.mcp.luxalgo.com/mcp, and under Authentication choose Required when the server asks. Claude connects anonymously, every keyless tool works right away, and the sign-in card only appears the first time you call an account tool. This Claude-only address exists because Claude.ai otherwise opens the OAuth window at connect time for any server whose OAuth metadata it can discover (claude-ai-mcp#1013); it serves the same tools with that metadata hidden until a sign-in is actually needed. Every other client uses https://mcp.luxalgo.com/mcp.

Claude Code

claude mcp add --transport http luxalgo https://mcp.luxalgo.com/mcp

Cursor

Use the Install in Cursor button above, or add this to .cursor/mcp.json:

{
  "mcpServers": {
    "luxalgo": {
      "url": "https://mcp.luxalgo.com/mcp"
    }
  }
}

Any other MCP client

Point your client's MCP config at the hosted URL:

{
  "mcpServers": {
    "luxalgo": {
      "url": "https://mcp.luxalgo.com/mcp"
    }
  }
}

Client	Where to add it
Cursor	.cursor/mcp.json, or the install button above
Claude Desktop	claude_desktop_config.json
VS Code	Install button above, or MCP settings
Windsurf	~/.codeium/windsurf/mcp_config.json
Zed	settings.json under context_servers
Warp	Settings → Agents → MCP servers
LM Studio	mcp.json
OpenCode	opencode.json
Gemini CLI	~/.gemini/settings.json

Local (stdio)

Runs every hosted tool locally, and unlocks the broker tools. Set read-only credential env vars for the brokers you use. Any subset works: a broker connects when all of its vars are set, and with no vars at all the broker tools simply stay unconfigured.

{
  "mcpServers": {
    "luxalgo": {
      "command": "npx",
      "args": ["-y", "@luxalgo/mcp"],
      "env": {
        "BROKERS_ALPACA_API_KEY": "…",
        "BROKERS_ALPACA_API_SECRET": "…",
        "BROKERS_KRAKEN_API_KEY": "…",
        "BROKERS_KRAKEN_API_SECRET": "…",
        "BROKERS_HYPERLIQUID_WALLET_ADDRESS": "0x…"
      }
    }
  }
}

Env var names derive from each broker's credential fields: BROKERS_<BROKER>_<FIELD> (for example BROKERS_OKX_PASSPHRASE, BROKERS_IBKR_FLEX_FLEX_TOKEN). The broker_setup tool lists every supported broker, its exact variables, and a one-line guide to creating each key with read-only scope, which is all this server ever needs.

Signing in with LuxAlgo (optional)

Almost everything here is keyless and works without an account. The tools in the Account and Trade Journal sections below need to know who you are; they use your LuxAlgo account through standard OAuth 2.1, with app.luxalgo.com as the authorization server. Nothing is required up front: every client can connect, list tools and use the public ones anonymously, and sign-in is only requested when you first call an account tool.

Hosted (ChatGPT, Claude, Cursor, any remote connector). The server advertises its protected-resource metadata and answers an unauthenticated account-tool call with a 401 + WWW-Authenticate challenge; MCP clients handle the rest (discovery, PKCE, consent screen in your browser) and keep the token for you. Each tool also declares its policy in tools/list (securitySchemes: noauth for public tools, oauth2 for account tools), so ChatGPT's per-tool linking works as well. Clients may identify themselves via Client ID Metadata Documents or Dynamic Client Registration — the app accepts both. One client-side exception: Claude.ai/Desktop connectors sign in at connect time whenever OAuth metadata is discoverable, regardless of their Authentication setting, so they get their own address, https://claude.mcp.luxalgo.com/mcp, whose metadata is only reachable from the 401 (see Install).

Local (stdio). The server running on your machine is itself the OAuth client. Sign in once:

npx -y @luxalgo/mcp login     # opens your browser; tokens are stored under your user config dir (0600)
npx -y @luxalgo/mcp status    # who is signed in, token expiry
npx -y @luxalgo/mcp logout

Tokens live in ~/.config/luxalgo/mcp-auth.json (%APPDATA%\luxalgo\mcp-auth.json on Windows, or LUXALGO_MCP_AUTH_FILE), are refreshed automatically, and are only ever sent to the LuxAlgo app. If your MCP client supports URL-mode elicitation (MCP 2026-07-28), you can skip the command: the first account-tool call asks the client to open the sign-in page and continues once you approve. Otherwise the tool answers with the challenge and the login hint.

What the token is for. This server never decides what you are entitled to — its code is public, so any such check would be decorative. Instead, once you are signed in, every request a tool makes to the LuxAlgo app carries your token, and the app resolves your account and plan from it exactly as it does when you use the web app. Public tools work without it; with it, the app can tailor what they return. When the app declines — no valid sign-in (401) or a feature outside your plan (403) — the tool reports that, naming the permission involved.

Tools

Library

Tool	Description
library_search	One search over concepts (alias-aware) and indicators
library_get_concept	Full concept page as markdown
library_get_indicator	Indicator detail: body, family, concepts, source code availability
library_get_source_code	Full source code when publicly served, fetched only on demand; otherwise chart_url, the indicator on a Vela chart
library_list_concepts	Paginated concept roster, optionally per family
library_list_indicators	Filtered, paginated browse (family, concept, tags, platform, tier) with server-side sort
library_list_tags	The indicator tag vocabulary, for the tags filter
library_list_families	The taxonomy backbone with counts
library_get_family	A family hub as markdown plus concept roster

Library outputs are compact JSON with canonical urls for citation. Concept and family pages are also directly fetchable as markdown: append .md to any concept URL.

Account (sign-in required)

Tool	Description
luxalgo_account	The signed-in user's plan tier, entitlements (alerts, historical bars, AI credits, …) and profile basics — so an agent can tailor answers to what the plan actually allows

Trade Journal (sign-in required)

Your own trade journal in the LuxAlgo app — the same accounts, trades, annotations and notes the app shows — read and written as you. Dates are YYYY-MM-DD day keys in your journal timezone (journal_list_accounts reports it); account filters take ids from the same call.

Tool	Description
journal_list_accounts	Journal accounts (broker-synced, imported or manual; currency, initial balance, lot method, last sync, archived state) and the journal timezone — the first call, since every accounts filter takes these ids
journal_overview	The dashboard for a window: performance metrics, Edge Score, per-day P&L, equity curve, open positions, accounts and settings; compare adds the previous equal-length window
journal_calendar	One month of the P&L calendar: day cells, weekly and monthly totals, trading and winning days
journal_breakdown	Closed trades grouped by weekday, time of day, hold time, symbol, side, position size, tag, rating and asset class — where the P&L comes from
journal_list_trades	Trade summaries, keyset-paginated; filter by accounts, open-day window, symbol, direction, status, tag; sort by opened/closed time, net or gross P&L, duration, quantity, symbol or rating, either direction
journal_get_trade	One trade in full: fills (reported values, corrections, hidden), per-exit P&L, every annotation
journal_get_day	A single day's stats, trades and notes
journal_list_tags	The user's annotation vocabulary — tags, mistakes, playbooks with usage counts — so new annotations reuse existing words
journal_search_notes	Day notes and trade notes as one newest-first stream; text query, day window, symbol and account filters, paginated
journal_add_trade	Log a trade by hand: its fills into a manual or import account; returns the resulting trade(s)
journal_update_trade	Annotate a trade: notes, tags (replace or add/remove), mistakes, playbook, rating, stop loss, profit target, reviewed
journal_write_note	Add a note to a day
journal_update_note	Replace a day note's text or move it to another day

The journal tools and luxalgo_account are the only tools that need a LuxAlgo account; see Signing in with LuxAlgo. Without a sign-in they return an OAuth challenge instead of data — never a silent fallback. The write tools act only as the signed-in user and only on that user's journal; the app validates and owns every change.

Brokers (local stdio only)

Tool	Description
broker_setup	Supported brokers, their env vars (set or unset, never values), read-only key guides
broker_accounts	Connected accounts: broker, currency, equity, cash
broker_positions	Open positions with market values, asset class, entry price; negative quantity means short
broker_trades	Trade history, newest first; filter by broker or symbol
broker_stats	Total equity, equity by broker, top positions, FIFO win rate and realized PnL
broker_refresh	Bypass the 5-minute cache and re-fetch now

Read-only by construction: the SDK's root export has no trading endpoints, the server never writes secrets anywhere, and per-broker failures are reported alongside results, never silently dropped.

Edge Stats

Hosted session statistics from the open-source edge-stats engine, precomputed nightly:

Tool	Description
edge_symbols	What the hosted store covers: symbols, session calendars, coverage windows, last build
edge_presets	The catalog of precomputed questions, each stating in plain language what its number means
edge_report	One result in the engine's full honesty envelope: estimate, N, Wilson 95% CI, minimum-sample guards, stability split, per-year counts, distribution, disclaimer

Every number arrives with its sample size — the engine has no code path that returns a percentage without one. Results are historical conditional frequencies, never predictions. For arbitrary composed queries or your own market data, run edge-stats locally; its own MCP server exposes the full engine over your local store.

Market Trackers

Tool	Description
trackers_datasets	The catalog: every dataset's row count, freshness, years with data (live vs deep-history archive), ticker-searchability; pass dataset for its field roster, filterable paths, caveats, per-year coverage, source health and dump URLs
trackers_query	Search one dataset by ticker, free text, exact field values (where, dot paths) and event-date range, choosing which years to read; newest/oldest ordering with paging
trackers_latest	The newest daily delta of a dataset (today's insider filings, this week's congressional disclosures), optionally narrowed by ticker or text
trackers_ticker	One ticker across every ticker-bearing dataset for a year: per-dataset counts with the newest rows — a public-record dossier

The data is the CC0 output of LuxAlgo/market-trackers, published daily to LuxAlgo/market-trackers-data: year-sharded gzipped JSON in the repository's live tree, plus deep-history shards attached to the data repo's GitHub Releases and indexed in its archives.json. The server streams shards row by row (never loading a whole file) under a per-call budget of compressed bytes, so a deep-history year (often 30–60 MB compressed) is read one at a time. Amounts disclosed as ranges stay ranges; ticker mappings for contracts, lobbying, trials, FDA events and patents are best-effort against a curated map of public companies; every row keeps its primary-source deep link.

Challenge Simulator

Runs locally inside the server:

Tool	Description
propfirms_list_simulatable	Every simulatable firm and challenge in the live directory, provenance-disclosed
propfirms_challenge_rules	One challenge's full encoded ruleset (drawdown modes, consistency, payout gating, citations), editable and re-simulatable inline
propfirms_simulate	Monte Carlo of your stats (win rate, avg win, trades/day, risk sizing) through a firm's exact ruleset and funded horizon: pass probability with CI, which rule kills attempts, expected attempts and cost, EV, payout probability
propfirms_simulate_trades	Same, from your real R-multiple trade series; block bootstrap preserves your streaks
propfirms_optimal_risk	Risk sweep: pass-optimal vs EV-optimal risk per trade (they differ)
propfirms_compare	Same trader across up to 12 challenges, EV-sorted (not a ranking)
propfirms_pass_rates	The site's reference-archetype odds, recomputed live (seed 42, 10k paths)
propfirms_validate_strategy	Screen one strategy across every simulatable challenge against an explicit pass bar

Every simulation result carries its assumptions, unsimulated-rule flags, seed, and engine version. Results are distributions under stated assumptions, never promises. The engine runs locally; firm rules adapt live from the directory, and inline specs simulate fully offline.

Prop Firm Directory

The live directory the simulator draws from, queryable directly:

Tool	Description
propfirms_search	Search firms; firm filters (platforms, markets, payments, Trustpilot, country availability) compose with nested challenge and offer filters, and include nests matching children
propfirms_get	One firm's full dossier: profile, every challenge, live offers, written overview
propfirms_search_challenges	Search challenges by rules (size, fee, steps, profit split, drawdown, trading restrictions) and parent firm; can attach applicable live offers
propfirms_search_offers	Current discounts and promo codes, resolvable per firm or per challenge

Charts, in your browser, with Vela

Every tool above returns text and JSON. When the answer wants a chart, draw it with the open-source core of Vela by LuxAlgo (@luxalgo/vela, Apache-2.0, GitHub), the charting library behind Vela: a headless chart with its own WebGL2 renderer that takes bars you already have, or fetches them from keyless public providers, and runs indicator scripts through pluggable engines. Pine Script® lives in the @luxalgo/vela-pinets addon, which is what closes the loop with the Library: library_get_source_code hands an agent an indicator's exact Pine source, and Vela's open-source core executes that source on a chart.

To open an indicator on Vela, LuxAlgo's next-generation charting platform at vela.luxalgo.com/chart, give the user the chart_url that library_get_source_code returns when the indicator's code runs only on LuxAlgo.

Not a mockup: the Library's SuperTrend source as returned by library_get_source_code, executed by @luxalgo/vela-pinets on a @luxalgo/vela 0.6 chart and screenshotted in headless Chromium. The bars are a labelled synthetic sample; point data at your own or register a provider for live ones.

The whole demo is two script tags and five lines. source is the source field of a library_get_source_code result:

<div id="chart" style="height: 480px"></div>
<script src="https://cdn.jsdelivr.net/npm/@luxalgo/vela@0.6.15/dist/vela.global.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/@luxalgo/vela-pinets@0.2.10/dist/vela-pinets.global.min.js"></script>
<script>
  const chart = new Vela.Vela('#chart', { data: bars, timeframe: '1D', theme: 'dark' }); // bars: [{ time, open, high, low, close, volume? }]
  chart.registerEngine('pine', new VelaPinets.PineEngine());
  chart.addIndicator(source);
</script>

With a bundler it is the same three calls over import { Vela } from '@luxalgo/vela' and import { PineEngine } from '@luxalgo/vela-pinets'; see Vela's quickstart. The same chart paints your own trades: Trade Journal takes the shape broker_trades returns and draws entries, exits and P&L labels through Vela's native-indicator API, engine-free, in one component you can lift as is.

Where each piece runs. This matters because it is the opposite of how the rest of this server works: