Seventeen typed tools
MCP names and schemas map directly to canonical Glade operations.
AI agents
Give agents tools, not unrestricted browsing.
Expose the same bounded Amazon operations through MCP and the Glade amazon-data Skill with one authorization model.
Why it matters
Context makes data usable.
Agents need typed inputs, predictable cost, freshness, and explicit failure states. Marketplace content must remain data, never instructions.
REQUESTproduct · one successful unit
POST /api/graphql
operation: product
variables: {
"asin": "B0D1XD1ZV3",
"domain": "US"
}Cost aware
Successful calls consume one unit and surface request metadata.
Injection resistant
Provider and marketplace content is returned as untrusted data.
Start with the contract, then verify with data.
Use the playground for a bounded live call, inspect freshness and request metadata, and move to production only after your fixture checks pass.
Open the playground