MCP server

BackQuant ships a remote Model Context Protocol server, so the same data behind the v2 API is available conversationally inside your own AI client.

https://mcp.backquant.com/mcp

https://api.backquant.com/mcp is the same server if your client prefers the API hostname. There are two ways in. Clients that speak OAuth (Claude, Cursor, VS Code) need nothing but your BackQuant login: sign in once, approve, done. Clients that only send a static header (the OpenAI Responses API, LangChain, n8n, your own scripts) use your existing API key as a bearer token. Either way your AI client calls the tools on your behalf and reads the answers back to you.

The server carries crypto only. Equities are not exposed through the API or MCP for upstream licensing reasons.

What you need

An active subscription that includes API access. That means any of:

Plan	Includes MCP
Crypto API (`api_only_monthly` / `api_only_yearly`)	Yes
Terminal + API	Yes
Terminal Yearly	Yes, at starter limits
Terminal Monthly	No, this plan has no API access

Your subscription must be active. A trial does not grant API access, so it does not grant MCP either. If you connect without an eligible plan the server tells you that during sign-in rather than failing later with a confusing tool error. For Claude, Cursor, VS Code and anything else that speaks OAuth. If your client only sends a static header, skip to clients that do not do OAuth instead.

Add the server to your client

Point your MCP client at https://mcp.backquant.com/mcp.In Claude (Desktop or Web), add it under Settings, Connectors, Add custom connector. In Cursor, add it as a remote MCP server. Any client that speaks Streamable HTTP with OAuth will work the same way.

Your client opens a BackQuant login page. Sign in with the account that holds your subscription and approve the connection.Clients that support dynamic client registration (RFC 7591) register themselves automatically, so there is nothing to configure by hand.

Ask something

Try “What are the BTC gamma levels right now?” or “Where did aggressors buy calls on ETH today?” The client picks the right tool and calls it.

Clients that do not do OAuth

Some clients never run the sign-in dance. They send one static header and expect that to be the whole story: the OpenAI Responses API, LangChain, n8n, and anything you write yourself. For those, pass your BackQuant API key as a bearer token. It is the same key you already create at the API dashboard, not a separate MCP credential, and it carries the same plan and quota.

curl https://mcp.backquant.com/mcp \
  -H "Authorization: Bearer bq_live_your_key_here" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

With the OpenAI Responses API, give the tool your MCP URL and put the same key in authorization. OpenAI does not perform the OAuth exchange itself, so the bearer path is the supported route there:

{
  "type": "mcp",
  "server_label": "backquant",
  "server_url": "https://mcp.backquant.com/mcp",
  "authorization": "bq_live_your_key_here"
}

Treat the key like any other secret. A bearer token is sent on every call, so prefer an environment variable over pasting it into a shared workflow, and rotate it from the dashboard if it is ever exposed.

Local clients that only speak stdio

Older MCP clients launch a local process and talk to it over stdin/stdout rather than calling a URL. Bridge them with mcp-remote, which runs locally and forwards to the server. It needs Node; there is nothing to install by hand, npx fetches it on first run. Add this to your client’s MCP config (claude_desktop_config.json for Claude Desktop, .cursor/mcp.json for Cursor):

{
  "mcpServers": {
    "backquant": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.backquant.com/mcp"]
    }
  }
}

That opens a browser once to sign in, then caches the token. To skip the browser entirely, hand the bridge your API key instead:

{
  "mcpServers": {
    "backquant": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://mcp.backquant.com/mcp",
        "--header", "Authorization: Bearer bq_live_your_key_here"
      ]
    }
  }
}

Restart the client after editing its config. If you already connect over a URL, you do not need any of this.

Tools

Nineteen tools, grouped by what you would ask for. Each returns a shaped, compact answer rather than a raw dump, so a question costs a sensible number of tokens. Positioning and levels

Tool	Answers
`get_gex_levels`	Gamma flip (HVL), call wall, put support for a market
`get_gex_strike_profile`	Dealer gamma broken down by strike
`get_greek`	Dealer exposure for a single greek, in several cuts
`get_max_pain`	The strike where the most option value expires worthless
`compare_symbols`	Gamma levels for several assets side by side, in one call

Flow and tape

Tool	Answers
`get_tape`	Individual prints and aggregates over them
`get_flow_by_strike`	Where aggressors actually bought and sold, strike by strike
`get_dealer_flow`	Dealer hedging pressure over time, measured from the tape

Volatility and pricing

Tool	Answers
`get_iv`	Implied volatility in whichever cut you need
`get_expected_move`	The move the market is pricing for an expiry
`get_probability`	Market-implied odds of finishing above or below a price
`get_basis`	The dated-futures curve, annualised basis by expiry

Structure and history

Tool	Answers
`get_oi`	Open interest by expiry or over time
`get_expiry_summary`	Every listed expiry with DTE, ATM IV and open interest
`get_opex`	Upcoming expiries and how much OI rolls off at each
`get_history`	How a positioning metric has changed over time
`get_liquidation`	Where leveraged perp positions get liquidated

Discovery

Tool	Answers
`list_symbols`	Which markets are covered and which have data right now
`list_expiries`	Valid expiry tokens to pass to the other tools

Limits

MCP runs on the same entitlements as the REST API. There is no separate MCP quota: a tool call is a request, counted against your plan’s monthly allowance and per-minute burst exactly like a REST call. See Rate limits. One question from a model often fires several tool calls in quick succession, which is why the starter burst allowance is set well above one call per question.

How it relates to the AI panel

They are complements, not substitutes.

	AI panel in the terminal	MCP
Requires	A terminal plan	An API plan
Runs	In the BackQuant terminal	In your own AI client
Data	Everything the terminal shows	Crypto v2 data
Sees your screen	Yes, it reads your panels	No

The panel gives depth inside our product. MCP puts our data inside your workflow.

Technical details

For anyone implementing a client rather than using one:


Transport	Streamable HTTP
Protocol revisions	`2025-06-18` (preferred), `2025-03-26`
Authorization	OAuth 2.1, PKCE required (`S256`), or an API key as a bearer token
Grants	`authorization_code`, `refresh_token`
Discovery	`/.well-known/oauth-authorization-server` and `/.well-known/oauth-protected-resource`
Client registration	Dynamic, RFC 7591, at `/oauth/register`
Issuer	`https://api.backquant.com`

Private-use redirect schemes (RFC 8252) are accepted, so desktop clients that use a custom scheme rather than a loopback URL connect without special-casing.

Tokens are scoped to the BackQuant account that approved them. Revoking the connection in your AI client, or the subscription lapsing, stops access at the next call.

MCP server

What you need

Clients that do not do OAuth

Local clients that only speak stdio

Tools

Limits

How it relates to the AI panel

Technical details

See also

What to ask it

Limits and errors

​What you need

​Connect with your login

​Clients that do not do OAuth

​Local clients that only speak stdio

​Tools

​Limits

​How it relates to the AI panel

​Technical details

​See also

What to ask it

Limits and errors

What you need

Connect with your login

Clients that do not do OAuth

Local clients that only speak stdio

Tools

Limits

How it relates to the AI panel

Technical details

See also