connect your ai agent

connect

https://rhxbt.com/mcp

streamable http transport, json responses, auth none. every tool returns observed data with nulls where nothing was verified, never fabricated zeros, and never a buy/sell recommendation. (/api/mcp/mcp still works as a legacy alias.)

claude code — one paste

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

or install the skill so claude knows when to reach for each tool: /skill.md — save it as .claude/skills/rhxbt/SKILL.md in your project.

claude desktop / cursor — one paste

claude desktop: settings → developer → edit config. cursor: .cursor/mcp.json:

{
  "mcpServers": {
    "rhxbt": {
      "type": "http",
      "url": "https://rhxbt.com/mcp"
    }
  }
}

chatgpt — one paste

settings → connectors → advanced → developer mode, then add a custom connector with this url (no authentication):

https://rhxbt.com/mcp

works in chatgpt's developer-mode connectors and deep research. any other client that speaks mcp streamable http connects the same way — one url, auth none.

pair it with robinhood agentic trading

robinhood's agentic trading mcp server places real orders — and carries zero chain data. connect BOTH servers to the same agent and the loop closes: risk-check here, trade there.

{
  "mcpServers": {
    "robinhood-trading": {
      "type": "http",
      "url": "https://agent.robinhood.com/mcp/trading"
    },
    "rhxbt": {
      "type": "http",
      "url": "https://rhxbt.com/mcp"
    }
  }
}

the guard rule (also in the skill): before any order on the trading side, call check_token_risk with the exact contract and trade size, and agent_signal_digest once per session. a "pass" gate means "nothing obviously wrong observed onchain" — never a recommendation. eligibility for the trading side is robinhood's (us accounts); this terminal is the data side.

three prompts to try

• "run the rhxbt signal digest and tell me if any stock token is dislocated more than 2% from its reference price right now"
• "check token risk for CASHCAT with a $500 position before i do anything, and show me which named checks failed"
• "any corporate actions pending on robinhood chain stocks this week? cross-check the oracle health for those tickers"

tools — when to call each

• agent_signal_digest — first call of every session: top premium dislocations + pending corporate actions + oracle anomalies in one compact read
• check_token_risk — before ANY trade: rug guard, named safety checks with proof links, radar risk gate, exit impact for your trade size
• get_stock_premiums — before pricing a stock token: dex vs live us print (session open) vs frozen oracle (closed), with the dated session verdict
• chainlink_feed — before trusting an oracle-referenced price: one feed's cadence, oraclePaused(), deviation, verdict
• pending_corporate_actions — before reading any stock chart: is a split scheduled that explains the move before it happens
• get_trending_tokens — finding ideas: trend scores 0-100 vs each token's own 7d baseline
• get_token — resolving a symbol safely: canonical stock contracts win over memecoin impersonators; full detail, signals, whales
• get_token_candles — charting: ohlcv 5m / 1h / 1d, the same bars the terminal draws, each with its source
• search_tokens — fuzzy lookup by symbol, name or address fragment with live market fields
• get_launch_radar — early candidates with the three-output assessment: risk gate / opportunity / confidence
• get_kol_leaderboard — who to listen to: callers ranked only by verified onchain call performance
• get_intel — what changed: oracle divergence, whale moves, depegs, volume spikes
• get_narratives — what the chain is talking about, ranked by mindshare with stability labels
• get_chain_stats — macro context: tvl, dex volume, txns, addresses
• get_equity_flows — how a stock actually settles: off-hours share, mint/burn, dvp prints per window
• get_settlement_graph — who trades stock tokens with whom off-pool — counterparty aggregates
• get_stock_multipliers — erc-8056 corporate actions already detected from the transfer ratio onchain
• stock_float — is real inventory being created: ap creations/redemptions per stock, net shares
• get_lending_markets — morpho markets with stock-collateral leverage flagged
• stablecoin_flows — the float behind everything: per-stable supply at par, changes, mints/burns, deployed ratio
• earn_tracker — what backs robinhood earn: usdg markets, utilization, reported apy, series
• perps_markets — lighter perps with the perp-vs-spot basis where the ticker is a stock we track
• perps_funding — hourly funding series for one perp market
• chain_economics — the business model measured: sequencer fees from sampled blocks, deploys, erc-4337 share
• chain_composition — equities chain or memecoin chain: value + holders per asset class, daily rwa share
• get_bridge_flow — net l1 bridge flow, valuation-clocked, withheld over fabricated zeros

limits

• mcp endpoint: ~30 requests/min per ip; rest routes carry their own windows, documented per route in the openapi spec
• read-only: no wallet, no orders, no signing — this server cannot move money
• everything is free today; the x-price field in the openapi spec is 0 on every route, and a paid lane, if ever, will appear at /.well-known/x402 first (a clearly-marked placeholder today)
• freshness fails closed: values past their observation ceiling are withheld with a dataNote — absent is absent, never zero
• no investment advice, ever. 0xrhXBT has NOT launched a token; verify at /token

machine surfaces

• /api/v2/openapi.json — the full rest surface: params, envelopes, rate limits, prices (all 0)
• /.well-known/agent.json — a2a-style agent card: skills, transport, auth none
• /.well-known/mcp.json — mcp registry descriptor
• /llms.txt — the machine-readable map of everything
• /skill.md — installable skill: when to call each tool + the trading pairing guard
• /datasets — every public series: schema, freshness, coverage, curl + mcp examples

also listed on smithery and the official mcp registry.