# CredCore Liquid

> An MCP connector over ~15,000 publicly filed credit agreements. Ask about
> covenants, baskets, pricing, and clause language in plain English, backed by
> ~240 analyst-curated datapoints across 25 sections. Published as a Claude
> connector, a ChatGPT plugin, a Gemini CLI extension, and a plain MCP server
> for any other MCP-compatible client.

Operated by CredCore. Informational only — not legal, lending, investment, or
underwriting advice.

## Connect

- Connector URL (streamable HTTP MCP): https://api.credcore.com/liquid-mcp/mcp
- Auth: OAuth 2.1 with PKCE and Dynamic Client Registration, scope `deals.read`.
  Protected-resource metadata (RFC 9728):
  https://api.credcore.com/.well-known/oauth-protected-resource/liquid-mcp/mcp
- Server card: https://liquid.credcore.com/.well-known/mcp/server-card.json
- API catalog (RFC 9727): https://liquid.credcore.com/.well-known/api-catalog
- An account is required. Sign up at https://liquid.credcore.com — access is
  granted from a waitlist, so a connection attempt without an approved account
  will fail at the authorization step.

## Tools

- `query_sql`: Execute a single read-only `SELECT` / `WITH … SELECT` over the
  deals database. Supports PostgreSQL regex helpers, a row cap, and offset
  pagination for large result sets.
- `search_values`: Full-text search across clause text, with optional section,
  datapoint, and deal-name filters. Returns ranked snippets.
- `lookup_columns`: Resolve a datapoint title to its column slug plus a
  plain-language definition of what the value means and how the stored cell is
  shaped (units, structure, conventions). Call this before designing a
  `query_sql` extraction.

All three are read-only, idempotent, non-destructive, and closed-world.

## Data

- `deals` — one row per credit agreement: `file_name`, `ticker`, `sector`,
  `sub_sector_1..3`, `document_year`, `document_date`, `ingested_at`.
- `datapoints` — metadata per (section, datapoint): `section_name`, `name`,
  `column_slug`. Datapoint ids are scoped per organization, not global.
- `facts` — long-format value store keyed by `(file_name, datapoint_id)` with
  `raw_markdown` values and a GIN-indexed full-text vector.

The corpus itself is access-controlled and is not published on this website.
The full datapoint glossary is reachable only through `lookup_columns` on the
authenticated connector.

Values are extracted from publicly filed credit agreements and reflect the
source document. Any analyst-style commentary (e.g. Lender-Friendly / Balanced
/ Borrower-Friendly verdicts) describes the source clause, not a recommendation.

## Docs

- [Connector setup guide (Markdown)](https://liquid.credcore.com/docs.md): full
  install instructions for Claude, ChatGPT, Gemini CLI, and other MCP clients.
- [Connector setup guide (HTML)](https://liquid.credcore.com/docs): same content
  as a web page.

## Optional

- [Privacy policy](https://www.iubenda.com/privacy-policy/20446622)
- Support: mcp-support@credcore.com