MCP Server

Endpoint: https://beliefsystems.xyz/api/mcp (MCP Streamable HTTP) Authentication: Authorization: Bearer belief_live_… – get a free key

As of August 29, 2026, the MCP server requires a free API key. Add an Authorization header to your client config – setup below. Tool names, arguments, and responses are unchanged.

The Belief Systems MCP server gives AI assistants and software agents structured access to the public data surface: the index catalog and per-index detail, chart-grade level history, the Belief Volatility series with the disclosures that must travel with them, and the point-in-time snapshot downloads.

The server mirrors the public data surface exactly. What sits outside it, and where that data lives instead, is in Data boundaries.

Setup

Create a key at beliefsystems.xyz/data. The Claude Code command takes the key from BELIEF_API_KEY in your shell and writes it into the client’s config, so rotating a key means re-running the command. The JSON config takes the secret directly, so paste it where belief_live_…your key… appears. The Codex command stores the variable name rather than the value and reads it at call time, so a rotated key is picked up by re-exporting it.

claude mcp add --transport http belief-systems \
  https://beliefsystems.xyz/api/mcp \
  --header "Authorization: Bearer $BELIEF_API_KEY"

{
  "mcpServers": {
    "belief-systems": {
      "type": "http",
      "url": "https://beliefsystems.xyz/api/mcp",
      "headers": {
        "Authorization": "Bearer belief_live_…your key…"
      }
    }
  }
}

codex mcp add belief-systems \
  --url https://beliefsystems.xyz/api/mcp \
  --bearer-token-env-var BELIEF_API_KEY

The portal’s MCP section carries the same snippets, and the dialog that shows a new key gives you one with the key already in it. If a tool call that worked yesterday now returns an error asking for a key, that is this change and not your configuration. Add the header. If you have a key but the error says it is not valid, it has been revoked or rotated – mint a new one in the portal.

Tools

Tool	Arguments	Returns
`list_indices`	–	The catalog of published indices: ticker, name, current level, inception date, publication status.
`get_index`	`ticker`	One index: level, inception, data quality, and links to the index page, a citable chart PNG, and downloads.
`get_index_history`	`ticker`, `horizon`	Chart-grade level history: up to 300 points of `{ t, indexLevel, stale }`, plus % change over the window.
`list_volatility_series`	–	The Belief Volatility catalog: ticker, name, headline value in probability points, the tenor the bare ticker denotes, publication state.
`get_volatility_index`	`ticker`	One Belief Volatility series with every disclosure that travels with it: release share per tenor, measure and tier labels, sensitivity band, frozen-prices counterfactual, per-leg breakdown.
`get_volatility_history`	`ticker`, `horizon`	Chart-grade Belief Volatility history with release shares and measure labels.
`get_data_downloads`	–	The current point-in-time snapshot: manifest, per-series CSV URLs, data dictionary, citation file.
`get_citation_guidance`	–	Attribution, licensing, and framing rules for quoting the data.

Argument details:

ticker – index or series ticker, e.g. CONFLICT, MIDR26, MIDVOL. Case-insensitive. Use list_indices or list_volatility_series for the catalogs.
horizon – one of 24h, 7d, 30d, all. Defaults to 30d. The windows match the citable chart PNG endpoint, so a quoted number and its chart always agree. all means since inception, clamped to the trailing 90-day free window; full history is in the snapshot downloads.

Every response carries the attribution line Source: Belief Systems (beliefsystems.xyz).

The MCP server now runs on the /v1 Data API. Tool names and schemas are unchanged; the one change existing clients need is the Authorization header.

Rate limits and fair use

60 requests per minute and 2,000 per day, per key. MCP calls and direct API calls draw on the same two windows and appear in the same usage record in the portal. Exceeding either returns an error quoting your limits and Retry-After. Full detail: Limits and errors.
A separate per-address bucket sits in front of the endpoint as an abuse shield, sized for interactive agent use rather than bulk collection. A client behaving normally never meets it.
History responses are chart-grade: capped at 300 downsampled points per call.

Scope and licensing

The endpoint is offered to registered account holders under a limited, revocable grant for evaluation, research, and internal business analytics – including use by AI assistants and agents acting on your behalf – with attribution. You may quote, cite, and display individual values in research, commentary, and editorial work, including externally published work. A separate written license covers:

redistribution
incorporation into a product, feed, benchmark, fund, or index-linked instrument
use in a production system relied on by you or a third party
extraction beyond the limits published for your keys

The exact language is in Terms of Service, Section 7. Keys are personal to the individual who registered the account; each person needing access registers separately. For live feeds, full-fidelity history, constituent detail, redistribution, or product use, see Index Services or request access.

Data API – the same data as a REST surface, on the same key
Point-in-time snapshots – per-series CSVs for research and editorial use
llms.txt – machine-readable site guide
Index Calculation Methodology – how index levels are computed

Getting Started

Understanding Indices

Belief Volatility

Transparency

Data Access

Risk & Legal

Reference

Setup

Tools

Rate limits and fair use

Scope and licensing

​Setup

​Tools

​Rate limits and fair use

​Scope and licensing

​Related surfaces

Setup

Tools

Rate limits and fair use

Scope and licensing

Related surfaces