mcpField
S
SpreadBench NCSX-IS — AI Infrastructure Bond Issuance Spread Index

SpreadBench NCSX-IS — AI Infrastructure Bond Issuance Spread Index

Remote
spreadbench.comFinance0Updated 1w ago

About SpreadBench NCSX-IS — AI Infrastructure Bond Issuance Spread Index

SpreadBench NCSX-IS is an MCP server for an event-driven issuance-spread benchmark covering AI-infrastructure bonds. It serves researchers and market analysts with bond, Treasury-curve, filing, and index data; its tools distinguish observed issuance data from withdrawn demo data and do not provide ratings or investment advice.

Key features

  1. 01Retrieve the latest index and two tiers
  2. 02Inspect 13 constituent bonds
  3. 03Look up issuer bonds and Treasury curves
  4. 04Review SEC filings and N-PORT marks
  5. 05Analyze maturity, events, and spread scenarios
  6. 06Read methodology and data-status fields

Use cases

  1. 01Compare secured and unsecured index levels
  2. 02Check an issuer's bond pricing spreads
  3. 03Trace bond events behind an index move
  4. 04Test a proposed coupon against peers

MCP support

Provides 14 read-only tools to MCP clients through a remote Streamable HTTP endpoint or a local stdio installation. Both transports use the same tool definitions and require no API key.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://spreadbench.com/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Tools

Details

From the project README

SpreadBench NCSX-IS — MCP server

Official Model Context Protocol server for SpreadBench NCSX-IS, an issuance-spread benchmark for AI-infrastructure (Neocloud) bonds.

What NCSX-IS measures. For each constituent bond, the coupon minus the US Treasury yield of matching tenor on that bond's pricing date, weighted by face value. It is a point-in-time issuance spread — event-driven, no daily frequency, and not a traded CDS level or a rating. SpreadBench publishes no credit ratings and no investment advice.

Live data: https://spreadbench.com · Methodology + DOI: https://spreadbench.com/methodology.html · JSON API: https://spreadbench.com/api.html


Two ways to connect

1. Remote (streamable-http, nothing to install)
{
  "mcpServers": {
    "ncsx": {
      "type": "streamable-http",
      "url": "https://spreadbench.com/mcp"
    }
  }
}

Free, no API key, no signup.

2. Local (stdio, same tool set)
npm install -g github:BrianTian818/spreadbench-ncsx-mcp
# or, without npm registry:
npm install -g https://spreadbench.com/download/ncsx-mcp-0.3.0.tgz
{
  "mcpServers": {
    "ncsx": { "command": "ncsx-mcp" }
  }
}

Both transports share one tool definition file (tools.js), so they never drift apart.


Tools

Fourteen read-only tools, all keyless. Every tool declares readOnlyHint: true / destructiveHint: false, so clients do not prompt for write permission.

ToolReturns
ncsx_latestStart here for "what is the level". Latest NCSX-IS reading, secured / unsecured sub-indices, constituent and issuer counts, notional, as-of date, methodology version, DOI, data_status
ncsx_indexStructural, not a single number. Full index definition, sub-index notionals, coverage notes
ncsx_constituentsThe whole universe, no filter. All 13 constituent bonds: issuer, ticker, coupon, maturity, size, seniority, pricing date, matched Treasury, issuance spread in bp
ncsx_bondFiltered lookup for one issuer — matches issuer / ticker / note keyword and returns only matching bonds
ncsx_curveThe Treasury par-yield curve used on the pricing date (14 tenors, linear interpolation, no extrapolation) plus recent history
ncsx_sectorUS high-yield / investment-grade OAS reference series for comparison
ncsx_filingsSEC filing radar for 9 AI-infrastructure issuers, last 45 days, noise forms filtered; pricing-chain forms (FWP / 424B* / S-3*) flagged pricing_signal
ncsx_marksSecond, independent layer: quarterly fair-value marks from SEC Form N-PORT filings, converted to spread
ncsx_wallMaturity wall by year with secured / unsecured split; pass year to drill into one cohort
ncsx_storiesCitable data stories recomputed live, each with headline numbers, bilingual page and a 1200x630 share card
ncsx_facts16 quotable numbers, each with value, unit, as_of, a human-readable calc and a source; filter by id or q
ncsx_eventsHistory as an event replay. 13 pricing events in order, level after each, change in bp and per-bond pull attribution (pull_bp), so pulls sum to the event delta; filter by since or ticker
ncsx_rulerReverse pricing. Given a coupon and a tenor: interpolated Treasury (never extrapolated), issuance spread, rank among the 13, gap to the index level and to the same-structure tier, and the index level after adding the bond at a given size. Pass issuer to get that issuer's own bonds already in the index as a pricing history. Recomputed on every call — nothing is stored. Also returns a like-for-like comparable set (peer_set: median / notional-weighted / range / rank inside the set; peer_window_years defaults to ±1y, and when the window is empty it falls back to the nearest tenor band with the distance stated).
ncsx_methodologyMethodology version, DOI, license, data sources and their licensing status, full formula chain

Three names from the demo-era server — ncsx_levels, ncsx_attribution, ncsx_weekly — are not listed any more, but still answer if an old client calls them: they return a structured retirement notice rather than an error.


Read data_status before you quote anything

Every payload carries a data_status field. This is the single most important field in the schema:

data_statusMeaning
real_issuanceObserved public issuance data. Safe to quote; reproducible from SEC filings.
retired_seed / seedA withdrawn demo series. Numbers were generated for layout testing. Never quote as market levels.

If a tool returns status: "retired", the endpoint was withdrawn — read the replacement field and move on. Nothing in this server will silently hand you demo numbers.


Example prompts

  • "What is the current NCSX-IS spread, and how does the unsecured cohort differ from the secured one?"
  • "Which CoreWeave bonds are in the index and what spread did each price at?"
  • "Compare NCSX-IS against the US high-yield OAS over the last 90 days."
  • "Which Treasury tenors does the index lean on most?"
  • "Is the NCSX-IS level consistent with the latest SEC N-PORT marks?"
  • "A $1.5bn 5-year note at 8.25% — where does that price relative to this market, and what would it do to NCSX-IS?"
  • "How did NCSX-IS get from 517.7bp to 432.3bp, and which issuer pulled it the most?"

Conventions

  • Identifier: constituents are identified by issuer + coupon + maturity. They do not carry CUSIPs — do not query by CUSIP.
  • Units: spreads in basis points over the Treasury at pricing. Notionals in USD millions.
  • Frequency: event-driven. There is no T+1 update and no daily change field.
  • Reproducibility: every number is derived from public filings (SEC EDGAR) and the US Treasury par-yield curve. The methodology page states the exact formula and the interpolation rule.

Notes on naming

Older versions of this server exposed Composite / HY / IG tiers and 90-day daily series. Those came from a demo dataset and were withdrawn on 2026-11-04 in favour of the issuance-spread methodology (v0.2.0-real). The three retired names listed above remain callable only so that old clients receive an explicit notice instead of an error; they are hidden from tools/list.

Directory listings

This server is listed in the official MCP Registry as com.spreadbench/ncsx, and is indexed by Glama (A tier) and brick.blue. glama.json in this repo claims the listing for its maintainer.

License

MIT. Data is published for research and citation; SpreadBench does not provide credit ratings or investment advice.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.