Amazon Agent Flow
Agents that know your Amazon business — not guess at it.
Your agents answer with real figures and take real actions on your Amazon business — across hundreds of accounts, from Claude, ChatGPT, or Cursor. It works because Kuudo lands your Selling Partner, Vendor, and Ads data in a private lake you own, then lets them act on it safely.
/the agent lake
Your Amazon data, in a private lake your agents can trust — on storage you own, ready to query from anywhere.
Kuudo turns your Amazon seller, vendor, and ads data into an AI-native Iceberg data layer on storage you own — so private, trusted agents can safely query the data they need to drive your business from DuckDB, Snowflake, Databricks, or any Iceberg-aware engine.
Apache Iceberg
The open table format every dataset lands in — schema evolution, snapshots, and time travel built in.
DuckDB
Query the lake in-process from a laptop or a function. No cluster to stand up, no warehouse to pay for.
Cloudflare R2
Your own zero-egress bucket. Bring it; Kuudo provisions and writes to it. The keys never leave your account.
Snowflake
Read the same Iceberg tables as Snowflake external tables — no copy, no reload, no second source of truth.
Databricks
Query the same Iceberg tables through Unity Catalog and Databricks SQL — no copy, right beside the rest of your lakehouse.
AWS Glue / Athena
Catalog the tables and run serverless SQL across the org — no servers to keep warm.
Amazon S3
Land Iceberg tables right next to the rest of your AWS stack, in a bucket you already control.
Your bucket, your catalog, your keys. Kuudo provisions the lake and keeps it fresh, but the Amazon seller, vendor, and ads data lands in infrastructure you control — ready for your private AI agents, not locked inside someone else's platform.
DuckDB runs in-process for repeated agent queries. Agent Flow estimates the data a governed query would scan and enforces your scan ceiling before execution. Large results stay server-side for analysis, keeping raw tables out of chat context.
Iceberg and Parquet keep your data usable everywhere: DuckDB on a laptop, Athena in AWS, Snowflake, Databricks, or the next engine your team adopts. Your agents get a trusted data layer without tying your business to one query stack.
Stable schemas, partition pruning, snapshot isolation, and access you control give AI agents the boring but critical foundation they need: reliable context, repeatable answers, and safe access to the Amazon data that drives your business.
/context for agents
Agents need context for smart, efficient queries. We deliver it.
Agent Flow tells your agents what data is available, what its fields contain, where it came from, and how fresh it is before they query. DuckDB runs the analysis, and compact results keep each follow-up focused on relevant evidence.
Context before the first query
Your agent gets dataset references, field definitions and classifications, freshness, and lineage before choosing what to query. It starts with information about the data already in hand, instead of spending queries trying to discover it.
Room to test and follow up
Agents work through questions in small steps: inspect, compare, check, refine. DuckDB runs in-process for that repeated analysis, and scan estimates let the agent size up a query before running it within your limits.
Keep the reasoning focused
Full results stay server-side for Python or curated analysis. The agent retrieves the relevant summary or small table for its next step, so its context stays focused on the evidence it needs to reason over.
You set the scan ceiling for Agent Flow’s governed queries. Agents can narrow it for a task; execution enforces your limit. The query-limit FAQ below explains how this differs from a spending cap.
MotherDuck’s analysis explains why context improves agent query accuracy and reduces the tokens needed to reach an answer. Read “Agents Don’t Query Like Humans Do.”
See the Agent Flow query and analysis workflow/taking actions
Tell it what to do. It does the work — change a bid, fix a listing, ship an order.
Your agent works inside your Amazon Seller, Vendor, and Ads accounts — not another read-only dashboard. You decide what each agent can touch, and every change it makes is logged.
Counts below are API operations in Kuudo’s published Amazon MCP inventories. Agent Flow exposes its enabled operation catalog separately through discovery.
Sellers
- get catalog data
- get ASIN matches
- create a listing
- update a listing
- delete a listing
- get orders
- confirm shipment
- update price
- update offers
- get fees
- create inbound plan
- update shipment
- get inventory
- create fulfillment
- cancel fulfillment
- update inventory
- create shipment
- confirm returns
- create inbound
- update shipment
- get inventory
- get rates
- create label
- confirm shipment
- submit feed
- get result
- request report
- get report
- run query
- get results
- get sales
- get metrics
- get settlements
- get financial events
- create transfer
- submit payout
- submit invoice
- create invoice
- create content
- submit content
- submit message
- create subscription
- submit review request
- get feedback
- create offer
- update offer
Vendor
- update inventory
- submit invoice
- confirm shipments
- update content
- update ASIN relations
- approve submissions
Ads
- create campaigns
- update bids
- create keywords
- create campaigns
- update creatives
- create campaigns
- update targeting
- create campaigns
- update creatives
- create campaigns
- update line items
- create campaigns
- update budgets & bids
- create targets
- update targets
- create creatives
- submit assets
- update store
- get analytics
- create audiences
- get insights
- create tags
- get attribution
- create conversion events
- get conversions
- request report
- get report
- get brand metrics
- get benchmarks
- get recommendations
- get insights
- create workflows
- create audiences
- run queries
- create subscription
- get stream
- get forecasts
- get reach
- get profiles
- create portfolio
- get billing
Most tools tell you a bid slipped or a listing broke. The agent goes and changes it — then confirms the work is done.
Run the same change across a single account or your whole book — agencies move every client at once, brands keep each marketplace in sync.
Decide what each agent can touch and what needs your sign-off. Every change is logged, so you always see exactly what it did.
/what kuudo runs
Get your Amazon data ready for agents — reliably, across every account.
Amazon data was not designed for agents. It sits in proprietary tools, scattered exports, brittle reports, and workflows that still depend on people copying sensitive business data into chat windows — or giving AI broad access to personal computers.
Kuudo changes the pattern. It pulls your Amazon seller, vendor, and ads data into one private place you control, then gives trusted AI agents safe ways to query, monitor, and act on it — without training the model to benefit your competitors.
Agents do the work
Kuudo is a native MCP server. Connect Claude or any MCP-compatible assistant and your agent can start an orders pull, request an ad report, check pipeline status, or query yesterday's numbers — all through approved actions and strict guardrails.
One connection to Amazon
Selling Partner. Amazon Ads. Orders, listings, inventory, catalog, reports, campaigns, performance, Sponsored Products / Display / Brands, and Reporting v3. Discover the available operations and inspect their inputs before running or scheduling a report.
Pipelines that don't drop data
Every run is durable. If Amazon throttles, a server restarts, or a job gets interrupted, Kuudo resumes from the exact step it left off — no duplicate loads, no silent gaps, no babysitting.
Built for hundreds of accounts
One deployment can run hundreds of seller and ad accounts, hundreds of schedules, and account-specific pacing that respects Amazon's limits. Hourly here, daily there, backfill when needed — all config-driven.
You decide when fixes run
The monitor ships observe-only. Review its audit log, then explicitly switch on automatic fixes. It can retry failed reads and recover pipeline work; it has no tools to change Amazon listings, prices, credentials, or accounts. Code enforces action caps, cooldowns, and attempt limits.
Healthy checks use zero model calls
Scheduled checks run for each tenant. Healthy cycles are recorded without calling a model; flagged anomalies go to AI triage. Every corrective action carries a recorded reason, so your team can review what happened and why.
Put AI to work on your Amazon business.
Connect Amazon Selling Partner and Amazon Ads, and your agents can finally act on your data — safely, starting with your own accounts.
Kuudo connects to Amazon Selling Partner and Amazon Ads. Capability availability may vary by connector and configuration.
Frequently asked questions