Skip to main content
Glama
@sentimentracker/signals-mcp

@sentimentracker/signals-mcp

by Aether-Grid

@sentimentracker/signals-mcp

A Model Context Protocol (MCP) server exposing Sentimentracker's 14 proprietary indicators as LLM-callable tools. Any MCP-aware client (Claude Desktop, Cursor, Cline, Continue, Windsurf, OpenAI Agents SDK) can ask "what is sniperV3 saying about BTC on 15m right now?" and get the actual signal envelope from your premium subscription.

Pairs with @sentimentracker/hyperliquid-mcp — that one signs and submits trades; this one reads signals.

How it works

Thin HTTP wrapper. Tool calls map directly to the existing Sentimentracker backend POST /superchart/indicators/calculate route, which is the single source of truth for every indicator's proprietary signal cascade. No calculator code lives in this package — backend updates ship to every MCP user immediately, no version drift.

LLM client → MCP stdio → this server → HTTPS → sentimentracker backend → cached metadata

Related MCP server: Data Puller MCP Server

Tools

Tool

Purpose

list_indicators

Live catalog from the backend registry — key, displayName, signalKind, description. Call once at session start.

get_indicator_signal

Latest signal envelope (signals[], trades[], currentTrade) for (indicator, symbol, resolution). ~5KB payload — uses metadataOnly=true. The everyday tool.

get_indicator_full

Full per-bar plot arrays + metadata. Typically 100-800 KB; anything over 1 MB of JSON is refused with RESULT_TOO_LARGE (use get_indicator_signal instead). Use only when the LLM needs per-bar context.

get_signal_confluence

Fan-fetch up to 20 indicators in parallel for the same (symbol, resolution), return a confluence table.

subscribe_signal

Long-poll for a NEW signal on (indicator, symbol, resolution). Holds up to timeoutMs (1-50s) and resolves the instant a signal newer than lastSeenSignalTs is detected.

propose_trade_from_signal

Pure compute: turn a signal envelope into a payload ready for @sentimentracker/hyperliquid-mcp's propose_order. Three sizing modes (notionalUsd / fixedBase / riskUsd) + optional TP/SL follow-up.

Supported indicators

The catalog is not shipped with this package. list_indicators reads GET /superchart/indicators on the backend, caches it for five minutes, and every other tool validates its indicator argument against that list. When the backend registers a new _specs/<key>.ts, it appears here on the next deploy with no MCP release. At the time of writing the backend serves 14 indicators (Sniper v3, Momentum Ultima Plus, Structure PRO, Strong Buy Strong Sell, MOM Algo V15, Institutional Algo, Momentum Trend Predictor, Momentum Radar, Momentum Wave, Momentum Balance Finder, Momentum Insider with Divergences, Momentum Levels V3, Divergences PRO, Impulse Targets).

Tool annotations follow the MCP spec: every tool is readOnlyHint and non destructive; propose_trade_from_signal is additionally openWorldHint: false because it never leaves the process. Results are returned as both text and structuredContent.

Auth

Two paths; both work today.

Env var

Status

When to use

SNT_SIGNALS_SESSION_TOKEN

Works today

Quick dev — log into Sentimentracker, copy session JWT, paste. Refresh on expiry (typically 7-30 days).

SNT_SIGNALS_API_KEY

Works (recommended)

Headless credential minted in Settings → API keys (or POST /api-keys). Premium only. Expires one year after minting; rotate any time from the same page (the old key keeps working for 24 hours). Use this for production / cron / agent loops.

Mint an API key from your account:

# Authenticate via your normal session JWT
curl -X POST https://api.sentimentracker.com/api/api-keys \
  -H "Authorization: Bearer <YOUR_SESSION_JWT>" \
  -H "Content-Type: application/json" \
  -d '{"name": "Claude Desktop — Mac", "scope": "signals:read"}'
# Response: { data: { key: {...}, plaintext: "snt_mcp_<24 chars>", warning: "..." } }

The plaintext field is shown once — capture and store securely. Lost keys can only be revoked + reminted, never recovered. List and revoke via GET /api-keys and DELETE /api-keys/:id.

The backend gates every indicator behind a premium subscription, so the JWT / API key must belong to a premium user. The MCP includes Authorization: Bearer <jwt> OR x-sentimentracker-api-key: snt_mcp_<...> automatically depending on which env var you set.

Install + run

From npm (requires Node.js):

SNT_SIGNALS_API_KEY=snt_mcp_... npx -y @sentimentracker/signals-mcp

Claude Desktop setup, with the config file location on macOS and Windows, is in INSTALL-CLAUDE-DESKTOP.md. Claude Code: claude mcp add sentimentracker-signals --env SNT_SIGNALS_API_KEY=snt_mcp_... -- npx -y @sentimentracker/signals-mcp.

Local development:

cd sentimentracker-signals-mcp
pnpm install
pnpm build
SNT_SIGNALS_SESSION_TOKEN=<your-jwt> node dist/index.js

Env reference

Env var

Default

Notes

SNT_SIGNALS_ENDPOINT

https://api.sentimentracker.com/api

Override for staging / local dev.

SNT_SIGNALS_SESSION_TOKEN

(required if no API key)

Bearer JWT from your authenticated session. Expires; refresh manually.

SNT_SIGNALS_API_KEY

(required if no session token)

Long-lived snt_mcp_… key minted via POST /api-keys. Recommended for headless / production.

SNT_SIGNALS_DEFAULT_PROVIDER

auto

One of auto / hyperliquid / binance / tiingo / intrinio / databento. Per-call override via tool input.

SNT_SIGNALS_REQUEST_TIMEOUT_MS

15000

Fetch timeout (1000-60000 ms).

Error envelope

Every failure returns a structured error:

{
	"code": "AUTH_PREMIUM_REQUIRED",
	"message": "Premium subscription required for indicators",
	"hint": "This indicator requires a Sentimentracker premium subscription.",
	"status": 403
}

Code

Meaning

Agent action

AUTH_MISSING

No session/key configured

Operator must set env var

AUTH_EXPIRED

401 from backend

Operator must refresh token

AUTH_PREMIUM_REQUIRED

403 from backend

Operator must upgrade subscription

BACKEND_TIMEOUT

Fetch timed out

Retry (sparingly) or raise SNT_SIGNALS_REQUEST_TIMEOUT_MS

BACKEND_HTTP_ERROR

5xx or other HTTP failure

Retry with backoff

BACKEND_INVALID_RESPONSE

Non-JSON or malformed body

Bug — file an issue

INPUT_UNKNOWN_INDICATOR

Tool was called with a key not in registry

Correct the input

INPUT_INVALID

Other input validation failure

Correct the input

Project layout

src/
├── index.ts              # stdio entry
├── server.ts             # McpServer factory
├── config.ts             # env parsing
├── client.ts             # HTTP client → /superchart/indicators/calculate
├── indicators.ts         # 14-indicator registry
├── errors.ts             # error hierarchy + toErrorPayload
└── tools/
    ├── _shared.ts        # defineTool wrapper + shared zod
    ├── index.ts          # registry
    ├── list-indicators.ts
    ├── get-indicator-signal.ts
    ├── get-indicator-full.ts
    ├── get-signal-confluence.ts
    ├── subscribe-signal.ts
    └── propose-trade-from-signal.ts

Rate limits and quota

The backend layers three limits on top of the global per-IP limit. All of them are per user, so the same user across machines shares one budget.

Limit

Value

Signal

POST /superchart/indicators/calculate

60 burst + 1/sec sustained (5× for premium)

X-RateLimit-Limit / -Remaining / -Reset (ISO 8601) on every response; 429 + Retry-After when exceeded

POST /superchart/indicators/subscribe

10 burst + 0.5/sec sustained (5× for premium); at most 3 open long polls per user

same headers; 429 when a 4th subscription is opened

API key monthly quota

50,000 successful calls per key per calendar month (UTC); session traffic is not metered

X-RateLimit-Monthly-Limit / -Remaining; 429 + Retry-After until the 1st

Surfaced to MCP callers as BackendError with code BACKEND_RATE_LIMITED and a Retry after Ns. hint.

Composing with @sentimentracker/hyperliquid-mcp

The two MCPs are designed to chain — signals MCP reads, hyperliquid MCP signs:

LLM
 ├── get_indicator_signal(sniperV3 BTC 15m)   → signals-mcp
 │     ↓ signal envelope
 ├── propose_trade_from_signal(...)            → signals-mcp (pure compute)
 │     ↓ HL order payload + sister MCP target
 ├── propose_order(payload)                    → hyperliquid-mcp (no signing)
 │     ↓ rendered payload + safety check
 └── place_order(payload)                      → hyperliquid-mcp (signs + submits)

propose_trade_from_signal emits a next.tool / next.server / next.arguments triple that tells the agent exactly which MCP server + tool to call next, so well-instructed agents handle the chain without explicit narration.

Roadmap

  • Real WS push (replace long-poll) — eliminates 1s latency floor on subscribe

  • Multi-symbol watchlists — one subscribe_watchlist covering N (indicator, symbol, resolution) tuples

  • Per-key analytics dashboard — call counts, error rates, top symbols

  • API-key scopes beyond signals:read (e.g. signals:subscribe, meta:read); scopes are enforced per route today

  • Hosted remote transport (Streamable HTTP + OAuth 2.1) alongside stdio

Releasing

  1. Bump version in package.json and SERVER_VERSION in src/version.ts (a test keeps them equal).

  2. Commit, then git tag vX.Y.Z and git push origin main vX.Y.Z.

  3. The publish workflow runs the tests and publishes to npm through trusted publishing, with provenance. No token or passkey is involved.

License

MIT

Available Tools

6 tools
get_indicator_fullGet full indicator outputA
Read-onlyIdempotent

FULL calculator output — every per-bar plot array AND the metadata envelope. Large payload (100-300 KB). Use only when the agent needs per-bar context; otherwise prefer get_indicator_signal. Pass from (unix seconds) to load history further back than the default window. Results over 1 MB of JSON are refused with RESULT_TOO_LARGE.

Inputs: symbol (required), indicator (required), resolution (required), from, +2 more.

get_indicator_signalGet latest signalA
Read-onlyIdempotent

Latest signal envelope for the given (indicator, symbol, resolution). Returns the proprietary metadata only — signals[], trades[], currentTrade — without the full per-bar plot arrays. Use this for 'what is sniperV3 saying about BTC on 15m right now?'. Optional config overrides the indicator's default params; provider forces a bar provider (auto / hyperliquid / binance / tiingo / intrinio / databento).

Inputs: symbol (required), indicator (required), resolution (required), config, +1 more.

get_signal_confluenceMulti indicator confluenceA
Read-onlyIdempotent

Run N indicators in parallel against the same (symbol, resolution) and return a compact confluence table. Each row: indicator key, displayName, signalKind, latest signal side (BUY/SELL/null), latest signal ts (ms), error (if any). The LLM judges agreement from the table — there's no built-in voting rule. Max 20 indicators per call; unknown keys fail the whole call before any fetch.

Inputs: symbol (required), indicators (required), resolution (required), provider.

list_indicatorsList indicatorsA
Read-onlyIdempotent

Catalog of every indicator the Sentimentracker backend can compute, read live from its registry. Returns key, displayName, signalKind (directional / regime / levels / divergence / mixed) and a one-line description per indicator. Call this once at session start so the agent knows which key to ask about.

Inputs: no named parameters.

propose_trade_from_signalCompose Hyperliquid order payloadA
Read-onlyIdempotent

PURE COMPUTE: compose a Hyperliquid propose_order payload from a signal envelope + sizing strategy. No IO. The result drops directly into @sentimentracker/hyperliquid-mcp's propose_order tool. Three sizing modes — notionalUsd / fixedBase / riskUsd (with stopLossPx). Optional takeProfitPx + stopLossPx surface a follow-up tip to invoke place_tpsl after the entry fills.

Inputs: coin (required), markPx (required), signal (required), sizing (required), +7 more.

subscribe_signalWait for next signalA
Read-onlyIdempotent

Long-poll for a NEW signal on (indicator, symbol, resolution). Holds the call for up to timeoutMs (1000-50000 ms) and resolves the instant a signal with time > lastSeenSignalTs is detected. Returns {signal: null, timedOut: true} if nothing happens within the window. Pair with get_indicator_signal to bootstrap lastSeenSignalTs. Idiomatic agent loop: get → subscribe → react → update watermark → loop.

Inputs: symbol (required), indicator (required), resolution (required), config, +3 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.5/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool maps to a distinct operation: catalog discovery, single-indicator signal envelope, full per-bar payload, multi-indicator confluence, long-poll subscription, and order composition. The two get_indicator_* tools are clearly separated by payload intent, and get_signal_confluence is explicitly a parallel multi-indicator view.

    Naming Consistency5/5

    All tool names follow a clean snake_case verb_noun pattern: list_indicators, get_indicator_signal, get_indicator_full, get_signal_confluence, subscribe_signal, propose_trade_from_signal. The shared get_ prefix and logical object naming make the API predictable.

    Tool Count5/5

    Six tools is well-scoped for a signals MCP: discovery, light read, heavy read, aggregation, streaming, and downstream trade proposal. No tool feels redundant or missing from the core workflow.

    Completeness5/5

    The tool surface covers the full signal lifecycle: enumerate indicators, fetch current signals, retrieve full history, compare confluence, subscribe to changes, and convert signals into trade payloads. There are no obvious dead ends or critical missing operations.

    Maintenance

    ActivityMaintained
    ResponsivenessOutdated data

    Related MCP Connectors

    • sigrexDeprecated
      io.sigrex

      Create and manage trading strategies, bots, webhooks and signals from any MCP client.

      https://mcp.sigrex.io
    • Crypto market signals, technical indicators, and sentiment analysis for AI agents.

      https://mcp.aarna.ai/mcp
    • > TradingMCP puts a market analyst inside your AI assistant. Paste one URL into Claude, ChatGPT, Cursor, VS Code or any MCP client and ask in plain language: screen for breakouts, run a multi-timeframe analysis, backtest a strategy with walk-forward validation, scan for smart-money and volume signals, or get a one-prompt morning market brief. In clients that support MCP Apps, price charts and the Markets panel are drawn right in the chat. > 40+ tools covering crypto (real-time), stocks, ETFs, FX, futures and commodities across major global exchanges. > Hosted, nothing to install: Pro $9/month (2,500 tool calls) or Pro+ $29/month (10,000). Paid upfront with a 14-day money-back guarantee; your API key arrives right after checkout. > Built on the open-source tradingview-mcp core (MIT, ~5k GitHub stars): self-host it free, or use the hosted version with zero ops. > For research and education, not financial advice. Not affiliated with TradingView

      https://mcp.cryptosieve.com/mcp
    • The full TraderMatrix market-data connector for Claude, Cursor and any MCP client. 51 tools: scored and tiered unusual options activity, dealer gamma (GEX), Apex levels and gamma pockets, IV rank, sector flow, dark-pool net-notional rollups, market health and macro regime, gap-continuation and breakout signals, wheel and CSP screeners, a one-call ticker snapshot, and your own watchlists. Needs a TraderMatrix API key (OAuth sign-in works for Claude web and mobile). A free, no-key connector with the gapper scanner and Market Health tools is listed separately as TraderMatrix Free.

    Related MCP Servers

    • F
      license
      Not graded
      quality
      C
      maintenance
      Enables MCP-compatible AI clients to access live crypto market data and AI-driven quantitative analysis, with structured outputs and full observability.
      -
    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables fetching live crypto market data from Binance, Bybit, and Hyperliquid via MCP tools, including klines, orderbook, funding rates, technical indicators, and market breadth for real-time analysis by LLMs.
      -
    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables LLMs to trade on MetaTrader 5 via REST API or MCP tools, supporting market/pending orders, position management, and account info retrieval.
      -
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI models to pull live market data, including quotes, technical indicators, charts, and TradingView screeners, through MCP tools.
      1
      MIT