MCP Tool Discovery
A client lists available tools and reads each tool’s name, description and input schema before calling it. KEYXE’s catalog is read-only and documents every tool’s status and error codes.
KEYXE MCP & AI
A Structured Interface for Human-Controlled AI Workflows.
Model Context Protocol can provide a consistent way to discover approved tools, request defined datasets and interpret structured results. KEYXE is being designed around transparent schemas, scoped permissions and clear separation of demonstration and production data access.
Design
Six ideas behind KEYXE’s MCP direction
MCP is KEYXE’s extensibility interface — important, but not a third Amazon API and not a way around Amazon’s data policies.
Query Inputs & Response Shapes
Inputs are validated against a schema — datasets, metrics and groupings come from allow-lists, and there is no free-form SQL. Responses are structured JSON with rows and metadata.
Scoped Organization Context
In production, each call is bound to one organization and checked against that organization’s connections and permissions. The demo uses a fictional “demo-org”.
Data Freshness
Every response carries an “as of” time, period, currency and warnings, so an AI client cannot mistake old or partial data for current totals.
Explicit Approval for Future Write Actions
No write tools exist today. Any future action tool would preview the change and require a person’s approval, separate from read access.
Third-Party Data Policy Gate
Transferring real Amazon data to an external AI provider is a separate decision governed by Amazon policies and disclosure duties. It is disabled in KEYXE unless a review and any required approvals allow it.
Example output
What a tool response looks like
A synthetic analytics_query result. Provenance travels with the data: mode, marketplace, currency, “as of” time and warnings. The flag is_live_amazon_data is always false in the demo.
The demonstration shows synthetic tool outputs only; external transmission of real Amazon data requires a separate permitted-use and disclosure assessment.
{
"mode": "synthetic_demo",
"tool": "analytics_query",
"dataset": "sample_ads_campaign_daily",
"marketplace": "US",
"currency": "USD",
"data_as_of": "2026-10-01T00:00:00Z",
"rows": [
{
"campaign": "Demo Campaign A",
"spend": 120,
"attributed_sales": 410
}
],
"row_count": 1,
"is_live_amazon_data": false,
"warnings": [
"Synthetic sample data. No Amazon API call was made."
]
}Planned architecture
How a future hosted MCP gateway would be gated
An MCP client would authenticate to KEYXE’s own authorization server (with PKCE, state and audience checks) and receive scopes for specific read-only tools. Amazon credentials are never passed to the client.
Each tool call would be evaluated against the organization, the connection, the dataset and the data-use policy — including whether the client counts as a third party under Amazon Ads policies.
Responses would carry source, marketplace, as-of time, units, currency, attribution notes, warnings and a request ID, and every call would be logged.
There is no public MCP endpoint today. mcp.keyxe.net will not be published until it is authorized and secured.
Run a tool in your browser.
The MCP Tool Explorer simulates read-only tools on synthetic data. Nothing leaves your browser.