Claude and Codex vs Adigy for KDP Amazon Ads

Claude or Codex can help manage Amazon Ads when connected to appropriate tools, permissions, and an execution system. MCP makes that connection easier. A working connector, however, supplies only part of what a KDP publisher needs to delegate advertising: a strategy, reliable data, recurring execution, spending limits, account history, and a way to recover from mistakes still have to come from somewhere.

Anthropic’s Claude and OpenAI’s Codex are the main examples in this comparison; OpenAI’s ChatGPT is also relevant when connected to suitable data and tools. Amazon’s official Ads MCP Server gives compatible AI applications access to advertising reports and campaign actions. Publisher Champ can expose publishing analytics to an assistant. Together, those capabilities create a credible starting point for a custom advertising agent.

Adigy offers a finished system for recurring book advertising management. Its purpose is to make the decisions and carry out the routine work across your selected books, formats, and markets while you spend your attention on publishing. The comparison is about how much of that operating system you want to build and maintain yourself.

Can Claude or Codex actually change Amazon Ads?

Yes, given a compatible connection, authorized account access, and the necessary tools. Model Context Protocol, or MCP, standardizes how AI applications connect to external data and tools. It can support a read-only analytics connection or a connection that performs actions. The permissions and capabilities of the particular server determine what is possible.

Amazon launched its Amazon Ads MCP Server in open beta on February 2, 2026. Its announcement describes campaign creation, updates and deletion, reporting queries, and workflows such as Sponsored Products campaign creation. Actual access depends on account eligibility, authorization, supported functionality, and the client you use. A model’s ability to understand an instruction does not establish that every advertising action is available in every account.

Amazon’s announcement also explains that “connectivity alone does not guarantee reliable outcomes.” Individual capabilities need coordination across an advertising workflow, with advertisers and partners contributing strategy, experience, and expertise. That is the useful starting point for evaluating dedicated software.

Publisher Champ supplies a different part of the picture

Publisher Champ brings together publishing sales, royalties, advertising spend, expenses, and related analytics. Its current Luna AI/MCP integration is described as read-only: an assistant can retrieve and analyze information, but cannot use those tools to edit Publisher Champ data or make changes to connected publishing accounts.

That can make an assistant useful for questions such as which series earned the most royalties after advertising over the last 90 days. Campaign changes require a separate execution connection. Adding Amazon Ads MCP supplies another capability; someone must still coordinate the publishing data, advertising actions, and decision rules.

What does each approach provide?

A useful comparison has four layers: reasoning, connectivity, an advertising management system, and publishing-specific economics. A general agent can participate in all four when someone engineers the surrounding application. MCP itself is a connection protocol; it does not prescribe the application’s strategy, scheduling, data storage, or recovery process.

Scroll horizontally on smaller screens to compare the approaches.

This does not establish that Adigy will outperform a well-engineered agent. It establishes which parts are supplied by the product and which parts the publisher must arrange. Read our KDP advertising software buying guide for the broader distinction between research, reporting, and execution.

Advertising management begins with data processing

Consider an illustrative catalog with 100 books, three marketplaces, six campaigns per book per marketplace, and 50 targets per campaign. That is 90,000 targeting entities before adding search terms, placement results, budgets, format variants, negatives, or historical observations. This is a scale example, not a claim about a typical Adigy customer.

The relevant information has relationships: a target belongs to a campaign, a campaign advertises a format in a marketplace, and several formats may belong to the same book. A strategy needs to preserve those relationships as data and campaigns change.

A database and ordinary code are well suited to grouping observations, calculating rolling spend and conversion rates, checking existing targets, and comparing periods. The calculations can be repeatable and inexpensive without asking a language model to reread every row on every cycle.

A capable general agent can use SQL tools and request only the summaries it needs. That is a sensible design. It also introduces an ingestion pipeline, a schema, an identity map, calculation code, and a query service. Those components become part of the custom advertising application you maintain.

Adigy starts with history and can retain it

According to Adigy’s founder, when a customer first connects an existing advertising account, Adigy can access up to the previous 15 months of available account data and save that history for future reference. A newly created Amazon Ads account naturally has no earlier activity to import. The history available depends on what the account and underlying data sources contain.

Keeping that information creates more context for future decisions: previous tests, seasonal patterns, format differences, and changes in performance need not disappear at the end of an assistant session. As a customer continues using Adigy, the retained history can grow.

More history can improve the evidence available to the system. It does not guarantee steadily improving results or mean old performance should override a changed market. Historical observations still need to be interpreted in context.

A same-day report can lead to the wrong decision

Suppose a target has spent $18 today and shows no sales. An author asks an agent to reduce its bid. The agent may carry out the instruction accurately, while the advertising decision remains premature.

Amazon’s advertising guidance for KDP explains why advertising and publishing reports can differ. Advertising reports show attributed activity; KDP reports cover overall book performance. Orders may take up to 12 hours to appear in advertising reports, and failed payments or canceled orders can subsequently change the figures. KENP and royalty reporting have their own behavior.

Sponsored Products reporting for authors uses a 14-day attribution window. That window describes when a purchase can be attributed to an earlier click; it does not mean every decision must wait exactly 14 days. It means recent observations can be incomplete and must be evaluated accordingly.

The decision may also depend on how much evidence has accumulated, the shopper’s intent, placement settings, format economics, and the amount the publisher is willing to spend testing. Lowering every bid after a few unsuccessful clicks can remove access to valuable auctions along with weak traffic. Our clicks without sales guide explains that distinction.

Define the publishing objective before optimizing

ACoS, ROAS, CPC, orders, and attributed sales are useful advertising measures. They do not individually tell you how much royalty income remains after advertising.

Adigy combines royalties and advertising spend across publishing dimensions such as book, format, and market. Royalties minus ad spend is advertising-adjusted royalty income. It is a useful operating measure, but it is not complete business profit when editing, production, software, tax, or other costs remain outside the calculation.

A general agent can calculate the same measure if it has reliable access to both sources and the correct mapping between them. Choosing what that measure should cause the system to do requires a policy too. See ACoS versus publishing profit and our royalty-based break-even calculator.

Use calculation code and AI for the work each handles well

Calculating spend, identifying duplicate targets, and enforcing a bid cap are defined operations. They belong in code that can be checked, repeated, and tested. Statistical estimates also need explicit assumptions and enough evidence; calculating an estimate precisely does not make its prediction certain.

Semantic relevance and buyer intent call for a different kind of judgment. A thriller about a retired naval officer may share vocabulary with naval history. Someone searching for “naval history books ww2,” however, may want nonfiction rather than a novel. Similar words do not establish the same purchase intent.

Adigy’s Negative Oracle evaluates fit using the book’s meaning and audience, alongside advertising evidence and the customer’s settings. The buyer and end reader may be different people, especially for gifts, children’s books, and activity books. Those distinctions help explain why generic string matching is inadequate.

A sensible architecture processes data, calculates evidence, evaluates semantic fit where needed, applies a policy, validates the resulting action, and records what happened. Embeddings, classifiers, language models, or a combination can participate in that design. Relationships can be cached and obvious mismatches filtered before expensive reasoning is requested. This describes useful engineering choices, rather than disclosing or asserting every detail of Adigy’s implementation.

Model errors become a control problem when money is involved

A model might confuse campaign spend with target spend, misread a marketplace, or infer a relationship the data does not establish. Conventional software can have bugs too. Every execution system needs checks appropriate to its actions.

For example, an AI component could propose a relevance score while code validates its format, checks the underlying book and target identities, applies the customer’s thresholds, and rejects any action outside spending or bid limits. The system should retain an action log and a recovery path. Connecting a tool should not give its reasoning component unrestricted authority over an account.

Adigy’s current controls include selected managed books, formats and markets, marketplace budgets, aggressiveness or target ACoS settings, negative review and reversal, protected keywords, and Master Undo. These controls support oversight and recovery; they do not make every decision infallible or undo money already spent on clicks.

The account needs a recurring operating loop

Amazon Marketing Stream is a push-based system providing hourly advertising metrics and near-real-time campaign information. Amazon describes uses including intraday bid optimization and reacting to budget consumption. Hourly metrics should not be confused with second-by-second, fully attributed sales data.

In an event-driven application, new information updates stored account state, triggers an evaluation, and may lead to a validated action. Other jobs run on a slower cadence for keyword discovery, search-term harvesting, and longer-horizon decisions. Persistent state lets the system remember its previous actions rather than start each analysis from scratch.

Adigy monitors continuously and makes hourly account updates. The Ads MCP announcement describes tools and workflows; connecting those tools alone does not establish a persistent subscriber, scheduler, or failure-monitoring service. A general agent can be scheduled and connected to streaming infrastructure, but those services must be configured and maintained.

Targeting itself is a loop: discover an opportunity, evaluate it, launch a test, collect evidence, promote useful terms, exclude poor fits, and discover again. Across a catalog, the system must also prevent conflicting positives and negatives and preserve the right book boundaries. A poor fit for one gardening book need not be a poor fit for another.

What would a comparable Claude or Codex system need?

A technical publisher can build a serious advertising application around a general agent. Codex supports MCP connections and can help write the surrounding software. Existing libraries and services can supply parts of the infrastructure; you do not have to invent everything from scratch.

The responsibilities still need an owner. A comparable system would need to address at least:

1. Persistent account state and historical snapshots.
2. Reliable recurring execution and job scheduling.
3. Advertising authentication, permissions, and access renewal.
4. Publishing royalty ingestion.
5. Book, ASIN, format, and marketplace reconciliation.
6. Delayed conversions and attribution handling.
7. Bid-optimization policies.
8. Placement optimization.
9. Campaign-budget allocation.
10. Target and search-term harvesting.
11. Negative-target logic and conflict prevention.
12. Campaign creation and expansion rules.
13. Strategies that vary by book, market, and format.
14. Account and marketplace spending constraints.
15. Retries, API errors, and prevention of duplicate actions.
16. Audit logs explaining what changed.
17. Checks that stop unexpected or excessive actions.
18. Recovery from unwanted changes.
19. Monitoring when the agent or its dependencies fail.
20. Validation when models, prompts, APIs, or Amazon behavior change.

At this point, the publisher is evaluating a custom advertising platform against a purchased one. Codex may be an excellent tool for building that platform. The coding assistant, its connectors, and the resulting operating application have different responsibilities.

Creative can change which targets make sense

Imagine a general “Would You Rather?” book for children. One Sponsored Brands video emphasizes a girls’ sleepover activity; another emphasizes a family car journey. Even though the advertised book is the same, the message and buying situation differ.

A useful targeting process can consider the book, the creative’s promise, the buyer and end reader, the candidate term, and the actual performance evidence together. Demographic cues alone should not determine relevance. A shopper needs a credible fit between the query, the ad, and what the book delivers.

A complete creative workflow could identify an audience, develop a concept, generate an asset, check Amazon requirements, create a campaign, select appropriate targets and negatives, set limits, and monitor the outcome. It would also need asset storage, provider integrations, moderation handling, and recovery from failed steps.

Adigy currently manages Sponsored Products and Sponsored Brands; automatic Sponsored Brands creative generation is planned. The end-to-end video example is a future workflow illustration, not a claim that Adigy already generates and deploys those videos today.

Practitioner experience compounds when it becomes software

Adigy’s founder, Ivan Finn, began working in KDP in 2017 and founded AdsDroid in 2020. Adigy started as an internal agency tool and became publicly available in 2024. The founder reports approximately $7 million in annual advertising spend managed across the agency team and software since 2020. That is a combined experience figure, not an audited claim about public-software spend alone.

The practical value of that background is the opportunity to turn repeated account problems into maintained operating logic. A situation is observed, an explanation investigated, a strategy evaluated, and an appropriate improvement encoded into the system. Not every observation should become a universal rule; book, market, and format differences matter.

This is our rationale for specialized software: the publisher can benefit from a maintained process without personally translating each new lesson into prompts, rules, and code. Experience is useful when it changes how the product handles real work. It does not guarantee a particular advertising outcome. Read our founder history and editorial methodology.

Where do Claude and Codex help most?

General agents are valuable for unusual questions, exploring hypotheses, external research, explaining a confusing report, drafting creative ideas, and prototyping custom workflows. They can offer flexibility beyond the questions anticipated by a product’s interface. The quality of an answer still depends on the model, tools, data, and instructions.

A publisher with engineering resources and a genuinely unusual strategy may reasonably build a custom system. That choice brings control and customization, together with responsibility for the operating loop. A publisher who mainly wants recurring advertising work handled may find a finished system a better use of time.

A specialized engine can work underneath a general assistant

A future arrangement could let an assistant explain a change in series profitability while a dedicated advertising engine maintains the account. The assistant might help a publisher decide whether to change a goal; the engine would execute recurring work within the approved boundaries.

An Adigy MCP/API interface enabling that arrangement is a possible direction discussed here, not an announced or currently available integration. General agents becoming more capable can make specialized, well-controlled services more useful to them.

Compare the whole operating system and its opportunity cost

We have not benchmarked a comparable autonomous Claude or Codex implementation, so we do not claim a specific cost advantage. A fair DIY budget would include model usage, data access, hosting, orchestration, monitoring, maintenance, and the publisher’s time, alongside any subscriptions.

Ask what you want to own. Building infrastructure can be worthwhile when it creates a distinctive business capability. If your goal is to publish more books and improve reader conversion, maintaining an advertising agent may compete with those priorities. Our opportunity-cost guide provides a framework for that decision.

MCP makes access easier. A dependable publishing advertising system still needs data, policies, execution, and accountability. Use AI where judgment helps, and checked software where the operation is defined. Adigy’s purpose is to supply that recurring management system so you can focus on the publishing business.

Explore Adigy’s current capabilities, review pricing, or talk through your catalog and advertising goals.