cpg-knowledge-graph

# CPG Knowledge Graph

> mcp.cpgknowledgegraph.ai — the data door of the GSC AI Agent Stack. GSC's SPARKS-classified product, brand, maker, retail and market intelligence, served to AI Agents over MCP.

The CPG Knowledge Graph is the Consumer Packaged Goods (CPG) graph that GreenCore Solutions Corp. (GSC) operates for Beauty & Personal Care (BPC) and CPG brands: makers on the supply side, retail grocery banners on the demand side, sovereign jurisdictional nodes under SM-ECO-10060 as the spine, GS1 GTINs as the identity, and ACM-68000 eligibility signals as the answer. It exists so a buyer's AI Agent can resolve a product, a brand, a maker or a banner deterministically before an order is placed, and so the answer is the same for every agent that asks.

SPARKS resolves six dimensions — SKU, Pack, Amount, Region, Kernel, and Standard — a CPG industry first.

GSC and the SPARKS standard are one operator: the graph that classifies the SKU and the standard the SKU is classified under are written, kept and served by GreenCore Solutions Corp.

Canonical endpoint: https://mcp.cpgknowledgegraph.ai/mcp (streamable-http). The server is stateful: `initialize` returns an `Mcp-Session-Id`, and every later call carries it.

This repository is a connect kit, not the deploy repo. It carries the tool list as the live server states it, the auth posture as the live auth.md states it, runnable examples against the live endpoint, and the Agent Skills in installable form. Nothing here carries a key, a secret or a customer.

## Tools

As returned by `tools/list` on the live server on 2026-09-09.

| Tool | What it answers |
| --- | --- |
| `resolve_gtin` | Resolve a GTIN to its procurement readiness signal. Zero-pads input to 14 digits. |
| `check_eligibility` | Check whether a GTIN can ship to a specific retailer banner in a country. |
| `get_signal_chain` | Return the full ACM-68000 signal chain for a GTIN. |
| `count_gtin_coverage` | Global coverage statistics: GTINs, nodes, makers (supply), retailers (demand). |
| `resolve_scope` | GSC Scoped Agents Registry — Era 1 founding registry plus the Era 2 station-keyed registry under SM-ECO-10060. |
| `list_nodes` | The sovereign nodes (the spine), each with its GS1 GTIN and per-node maker and retailer counts. |
| `resolve_node` | Resolve one node: its GS1 license, its makers (supply) and its retailers (demand). |
| `find_makers` | Global cross-node maker (supply) search by segment, region, node or website. |
| `find_retailers` | Global cross-node retailer (demand) search by region, node or parent banner. |
| `node_market` | Supply-to-demand bridge for one jurisdiction: makers and retailers as two separate counts, plus the same-node bridge pairs. |
| `list_sku_types` | SPARKS taxonomy: the standing BPC product-type codes the graph is structured on. |
| `list_brands_by_node` | SPARKS depth surface: the distinct brands SPARKS covers in a sovereign market. Coverage only. |
| `get_kernel` | SPARKS completeness tier for a brand in a market. Returns the stored tier flag, never a pack/size. |
| `resolve_sparks` | Two-branch SPARKS resolver: a GSC territory or license GTIN returns its record; any third-party BPC GTIN returns a gated proof, never the pack/size value. |

The graph changes; the live `tools/list` is the source of record for this table.

## Auth

As https://mcp.cpgknowledgegraph.ai/auth.md states it: public content and machine files on this surface are open by design, and the MCP endpoint serves its read tools openly. Transactional agentic commerce is authorized at the GSC Navigator boundary using Microsoft Entra OAuth 2.0, human-in-the-loop on every commit: reads are autonomous, every transactional commit is human-signed. The declared enforcement posture is open-by-declaration; when the buy side's credential standard reaches general availability, expected Q4 2026 to Q1 2027, transactional methods migrate to an OAuth 2.1 challenge per the MCP 2025-11-25 specification and read access stays open. Resource scope: `mcp:read`. Protected Resource Metadata at `/.well-known/oauth-protected-resource`.

## Examples

| Walk | Python | TypeScript |
| --- | --- | --- |
| resolve a GTIN (990832300082): initialize → `resolve_gtin` | `examples/kg_resolve.py` | `examples/kg_resolve.ts` |
| walk the six SPARKS dimensions for that GTIN: the dimension definitions from the standards beacon, then the graph tool that answers each one | `examples/sparks_walk.py` | `examples/sparks_walk.ts` |
| one banner/retailer lookup: `find_retailers` in one node, then `check_eligibility` for the GTIN at one banner | `examples/retailer_lookup.py` | `examples/retailer_lookup.ts` |

Python 3 standard library only. TypeScript runs on Node 24 as-is (`node examples/kg_resolve.ts`) with no dependencies. Each script prints what it got and stops. Live endpoints only.

## Skills

`skills/` holds the GSC CPG Agent Skills in the Agent Skills form: one skill per folder, a `SKILL.md` with `name` and `description` frontmatter, the live endpoint at mcp.cpgknowledgegraph.ai as the tool each skill calls.

| Skill | Folder |
| --- | --- |
| `gsc-cpg` — navigate the CPG Knowledge Graph: makers, banners, nodes, GTINs, eligibility signals, the SPARKS dimensions | `skills/gsc-cpg/` |
| `gsc-cpg-onboarding` — the path that makes a CPG product AI-orderable through the GSC stack | `skills/gsc-cpg-onboarding/` |

Install by path:

- Claude Code — copy a skill folder into `.claude/skills/` (project) or `~/.claude/skills/` (personal); or add this repository as a plugin marketplace (`/plugin marketplace add greencore-solutions/cpg-knowledge-graph`, then `/plugin install gsc-cpg@gsc-cpg`; the manifest is `.claude-plugin/`).
- GitHub Copilot — copy a skill folder into `.github/skills/` in the repository where the agent runs.
- Any Agent Skills client — the folder is the unit; `SKILL.md` is the entry.

Release: the Skills release tag points at this folder. Registry listing: `io.github.greencore-solutions/cpg-knowledge-graph`.

## Sources on the wire

https://cpgknowledgegraph.ai · https://mcp.cpgknowledgegraph.ai/auth.md · https://mcp.cpgknowledgegraph.ai/.well-known/agent-card.json · https://mcp.cpgknowledgegraph.ai/resolve?gtin=990832300082 · https://acm-sparks.ai · https://sm-eco-10060.org · https://acm-68000.org

---

GreenCore Solutions Corp. — 4611 Viking Way, Suite #260, Richmond BC V6V 2K9 Canada
GSC Agentic Pty Ltd — Level 1, 63-73 Ann Street, Surry Hills NSW 2010, Australia

Operated by GreenCore Solutions Corp. · https://gsc-em.com · @GSC_Rail_ai · @ACM68000