One Snowflake MCP endpoint for Claude, ChatGPT, and Gemini Enterprise.
Serve Claude, ChatGPT, and Gemini from a single JSON-RPC endpoint backed by Snowflake views, without exposing source credentials to client models.
Why Model Context Protocol (MCP)
AI models need live operational context to produce accurate, defensible work. Connecting enterprise systems to them has typically meant bespoke API wrappers, hand-built prompt pipelines, or unmonitored database exports.
Customer Zero uses the open Model Context Protocol (MCP), a JSON-RPC interface through which models discover tools, inspect schemas, and query resources. With Snowflake behind an MCP gateway, models query structured tables through token-authenticated views with logged permissions.
Governance boundaries & grant model
- Instance-scoped grants
- Consumer user × Source instance × Client
A query cannot cross a source boundary without an explicit, active grant record. Each grant names the consumer user, the exact source instance, and the authorised AI client runtime.
- Credential isolation
- No plaintext secrets in MCP context
Source credentials sit in Google Secret Manager as opaque resource pointers. The MCP gateway runs queries against Snowflake views, so clients never see or receive source tokens.
- Immutable audit trail
- Append-only audit events
Every MCP query, authorisation check, grant creation, and revocation emits an immutable audit event. Operators can review the full authorisation history through GET /audit-events.
Connect Claude, ChatGPT, or Gemini Enterprise
One MCP gateway serves every supported client. Operators provision a single MCP URL with token-based access, and each team uses the model runtime it prefers. The URL and token are issued after sign-in.
Claude
ChatGPT
Gemini Enterprise
Claude
- Open Settings, then Connectors, and choose Add custom connector.
- Paste your organisation's MCP server URL from the Customer Zero dashboard.
- Authorise with the access token issued to your consumer user.
ChatGPT Enterprise
- In workspace settings, enable developer mode and add a custom MCP connector.
- Enter the MCP server URL and choose token-based authentication.
- Paste the access token, then enable the connector in a chat.
Gemini Enterprise
- In the admin console, register a new MCP data source for your organisation.
- Provide the MCP server URL and the access token as the credential.
- Assign the data source to the teams that hold a matching grant.
Supported source catalogue & ingestion status
Ten business systems have a connection path into the Snowflake schema. Xero is connect-only for now, and Quinos POS and Raptor POS are in beta.
| Source | Connection model | Landing and models | Snowflake schema representation |
|---|---|---|---|
| Google Ads | Self-serve OAuth | Meltano tap, GCS JSONL, c0_raw_feeds then c0_analytics | Campaign, ad group, keyword, search-term, and spend tables |
| Meta Ads | Self-serve OAuth | Meltano tap, GCS JSONL, c0_raw_feeds then c0_analytics | Ad account, campaign, ad set, and custom audience spend tables |
| Google Analytics 4 | Self-serve OAuth | Meltano tap, GCS JSONL, c0_raw_feeds then c0_analytics | Traffic acquisition, event counts, page views, and conversion tables |
| Google Search Console | Self-serve OAuth | Meltano tap, GCS JSONL, c0_raw_feeds then c0_analytics | Search performance data for each verified site property |
| Shopify | Self-serve API token | Meltano tap, GCS JSONL, c0_raw_feeds then c0_analytics | Store data from the Shopify GraphQL Admin API, one store per instance |
| Clockify | Self-serve API key | Meltano tap, GCS JSONL, c0_raw_feeds then c0_analytics | Time entries, project allocations, billable rates, and member activity |
| Google Business Profile | Operator-provisioned | Meltano tap, GCS JSONL, c0_raw_feeds then c0_analytics | Per-location search impressions, review sentiment, and ratings |
| Quinos POS | Operator-provisioned agent (beta) | On-premise agent, GCS landing, c0_raw_feeds | On-premise point-of-sale line items, transactions, and tender types |
| Raptor POS | Operator-provisioned agent (beta) | Site-installed agent, GCS landing | Point-of-sale sales data from the site's Raptor server |
| Xero | Connect-only setup | None: no ingestion workflow runs | Accounting setup available; ingestion workflow not currently active |
Knowledge Harness & operational context
Numbers without context get misread. A 20% decline in marketing spend can look like a performance problem unless the model knows a seasonal promotion has ended.
The Customer Zero Knowledge Harness pairs quantitative Snowflake tables with qualitative operational documents: markdown playbooks, brand voice rules, campaign guidelines, and audience definitions. When an AI client queries the MCP endpoint, the Knowledge Harness returns the relevant business context alongside the metric rows, so answers reflect how the business operates.
Hosting and operations
- Operator
- IKMJ (Ikigai & Majime Pty Ltd)
An active Australian private company that builds and operates Customer Zero. ikigaimajime.com.au
- Hosting
- Google Cloud Platform, Melbourne, Victoria
The entire production MCP stack runs in this region. Ingestion pipelines, credential references, and Snowflake endpoints all run in Australia, with no cross-region replication.
- Certification
- None asserted
Customer Zero does not currently hold SOC 2, ISO 27001, or IRAP certification.
Technical teams can check the security controls and data handling on the Security & Trust page or request architecture documentation at sales@customerzero.app.