MCP App Store
by KeenEthics
Awin icon

Awin MCP Server for ChatGPT

By Awin

Overview

Connect your Awin publisher, advertiser, or agency accounts with your Awin login. Ask about live account access, performance, programmes, advertiser partnerships, help articles, and tracking links. Results and available actions are limited to the accounts and role you already have in Awin. No API token is needed. Depending on your role, you can also create tracking links and update publisher partnership tags; the assistant should show tag changes and ask you to confirm before applying them. Performance reports cover up to 31 days per request, and advertiser reporting may require API access on your Awin plan. The MCP provides a curated set of Awin capabilities rather than every platform screen.

How to connect

  1. ChatGPT: Open the Awin listing in the ChatGPT app directory and choose Connect. ChatGPT listing Awin exposes 22 tools here.

Tools

awin_compare_accounts

ChatGPT
Ranked snapshot of which accounts are doing well or badly. Uses only GET /accounts for the signed-in user (plus one report per selected account: advertiser publisher report or publisher advertiser report). Filters: account_type, account_ids, account_names. start_date and end_date are yyyy-MM-dd and at most 31 days; a longer range returns date_range_exceeded. APGW rate limits use separate buckets from GET /accounts: up to 100 publisher reports/min (100/min bucket) and up to 299 advertiser reports/min (300/min default bucket), counted per type. Larger selections return rate_limit_budget instead of a silent partial. Ranking is within account type and currency (commission, then transactions, then clicks) — do not mix types or currencies into one league table without labels. Every account row carries currency; money is never summed across currencies. Docs: https://help.awin.com/apidocs/get-publisher-performance-report and https://help.awin.com/apidocs/get-advertiser-performance-report

awin_get_account_performance

ChatGPT
Advertiser account health (GET /advertisers/{id}/reports/campaign then GET /advertisers/{id}/transactions/). Requires advertiser_id from awin_list_accounts. start_date and end_date are yyyy-MM-dd. Does not invent totals. If either call fails, ok is false and error, message, and incomplete mark the result as not the full account; do not present the successful half as the whole account. limit only pages the response: the full transaction list for the window is fetched from APGW on every call (no row limit is sent upstream), so the date range, not limit, sets the cost of a call. Windows are at most 31 days; a longer range returns date_range_exceeded without calling APGW -- shorten the range or query consecutive windows. An advertiser_id that is not in GET /accounts returns account_not_accessible and does not call the report: say they don't have access to this account and should use an account id from their account list, not that the plan lacks API access. A 403 on an id that is listed means this account's plan is not entitled to that API, not that there is no data. Campaign includes rows without a campaign parameter by default. Docs: https://help.awin.com/apidocs/get-campaign-data-for-advertiser-report and https://help.awin.com/apidocs/returns-a-list-of-transactions-for-a-given-advertiser If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_get_advertiser_performance

ChatGPT
Publisher advertiser performance report (GET /publishers/{id}/reports/advertiser): clicks, impressions, sales, and commission by advertiser for a date range. Requires publisher_id from awin_list_accounts. start_date and end_date are yyyy-MM-dd and at most 31 days; a longer range returns date_range_exceeded and is not estimated. A publisher_id that is not in GET /accounts returns account_not_accessible and does not call the report: say they don't have access to this account and should use an account id from their account list. A 403 on an id that is listed is entitlement_denied: the plan has no API access, not empty data. Does not include programme KPIs such as EPC — use awin_get_programme_details for those. Docs: https://help.awin.com/apidocs/get-advertiser-performance-report If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_get_advertiser_terms

ChatGPT
Returns the programme terms of up to 50 advertisers for a publisher account: commission rules (VAT, delivery, excluded categories, decline reasons), paid search and brand bidding rules, allowed partner types, deduplication, and additional and branding terms (HTML). A partnership is not required. Pass a publisher_id from awin_list_accounts and the advertiser ids to check. Advertisers that do not exist or have no stored terms are left out of results. Errors: account_not_accessible means the user cannot use this endpoint on the publisher (not linked, role too low or not entitled), never empty data; obo_exchange_refused and obo_exchange_failed mean the request could not be authorised. Report these and do not retry in this turn. If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_get_creative_performance

ChatGPT
Advertiser creative performance (GET /advertisers/{id}/reports/creative): clicks, impressions, and commission by creative for a date range. Requires advertiser_id from awin_list_accounts. start_date and end_date are yyyy-MM-dd and at most 31 days; a longer range returns date_range_exceeded without calling APGW. An advertiser_id that is not in GET /accounts returns account_not_accessible and does not call the report: say they don't have access to this account and should use an account id from their account list. A 403 on an id that is listed is entitlement_denied: the plan has no API access, not empty data. region is required by APGW 1.0 (default GB). Docs: https://help.awin.com/apidocs/get-advertiser-creative-performance-report If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_get_partner_performance

ChatGPT
Advertiser partner (publisher) performance (GET /advertisers/{id}/reports/publisher then GET /advertisers/{id}/publishers). Requires advertiser_id from awin_list_accounts. start_date and end_date are yyyy-MM-dd and at most 31 days; a longer range returns date_range_exceeded without calling APGW. An advertiser_id that is not in GET /accounts returns account_not_accessible and does not call the report: say they don't have access to this account and should use an account id from their account list. A 403 on an id that is listed is entitlement_denied: the plan has no API access, not empty data. If either call fails, ok is false and error, message, and incomplete mark the result as not the full report; do not present the report rows as complete when the publisher list failed, or vice versa. Join report rows to publisher info by publisher id on the caller side. Does not invent metrics. When the result is bounded, the report and publisher lists are each paged with the same limit. Docs: https://help.awin.com/apidocs/get-publisher-performance-report and https://help.awin.com/apidocs/get-publishers-information-for-advertiser If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_get_portfolio_kpis

ChatGPT
KPI table for all, advertisers-only, publishers-only, or named accounts. Uses only GET /accounts for the signed-in user plus one report per selected account (advertiser publisher report or publisher advertiser report). Filters: account_type, account_ids, account_names. start_date and end_date are yyyy-MM-dd and at most 31 days; a longer range returns date_range_exceeded. APGW rate limits use separate buckets from GET /accounts: up to 100 publisher reports/min (100/min bucket) and up to 299 advertiser reports/min (300/min default bucket), counted per type. Larger selections return rate_limit_budget instead of a silent partial. Totals are included only when every selected account succeeded and none was truncated. Every account row carries currency; money is never summed across currencies — totals give commission_by_currency and sale_amount_by_currency, and a single commission/sale_amount only when one currency is present. Docs: https://help.awin.com/apidocs/get-publisher-performance-report and https://help.awin.com/apidocs/get-advertiser-performance-report

awin_get_programme_details

ChatGPT
Get KPIs and membership for one programme (GET /publishers/{id}/programmedetails). Requires publisher_id from awin_list_accounts and advertiser_id (programme id). Returns EPC, conversion rate, approval percentage, and membershipStatus. Does not return a date-range performance report. Docs: https://help.awin.com/apidocs/get-program-information-details-for-publisher

awin_get_publisher_partnership_tags

ChatGPT
Returns the partnership tags an advertiser account has given to one publisher, ordered by name. Pass the advertiser_id from awin_list_accounts and the partner_publisher_id of a publisher the advertiser has a partnership with. A publisher with no tags returns an empty list; a publisher that does not exist or has no partnership with the advertiser is an error. Errors: account_not_accessible means the user cannot use this endpoint on the advertiser (not linked, role too low or not entitled), never empty data; obo_exchange_refused and obo_exchange_failed mean the request could not be authorised. Report these and do not retry in this turn. If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_list_accounts

ChatGPT
List every publisher and advertiser account for the signed-in user (GET /accounts). Takes no arguments and always returns all accounts, so one call gives the complete picture. Use the account_id values as publisher_id or advertiser_id on other tools. Does not return performance data. Also returns which Awin tools apply to each account: account_tools maps every account id to the tool names for its account type, and tool_catalog describes those tools once per type. Whether the signed-in user may actually call a given tool is enforced by Awin, not reported here.

awin_list_advertiser_partnerships

ChatGPT
Lists the publishers an advertiser account has a partnership with, with each partnership's status and when it reached it. partnership_status defaults to ACTIVE (joined publishers); pass other statuses, e.g. INVITE_PENDING for invitations the advertiser has sent, or APPLICATION_PENDING for applications waiting on the advertiser. sort is -partnership.statusUpdatedAt (newest first) unless given. Publishers whose account is not active are not listed. Pass an advertiser_id from awin_list_accounts. Errors: account_not_accessible means the user cannot use this endpoint on the advertiser (not linked, role too low or not entitled), never empty data; obo_exchange_refused and obo_exchange_failed mean the request could not be authorised. Report these and do not retry in this turn. If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_list_partnership_tags

ChatGPT
Lists the partnership tags an advertiser account has defined to group its publishers, ordered by name, each with the ids of the publishers that carry it (empty when none does). Use it to see how the advertiser segments its partners, for example before asking about the publishers behind a tag. Pass an advertiser_id from awin_list_accounts. On large accounts the result is paged and each tag's publisher list can be shortened: when nested_truncated is true those publisher lists are partial, so never report their length as the number of publishers; use awin_get_publisher_partnership_tags for a specific publisher. Errors: account_not_accessible means the user cannot use this endpoint on the advertiser (not linked, role too low or not entitled), never empty data; obo_exchange_refused and obo_exchange_failed mean the request could not be authorised. Report these and do not retry in this turn. If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_list_partnerships

ChatGPT
Lists the advertisers a publisher account has a partnership with, with each partnership's status and when it reached it. partnership_status defaults to ACTIVE (joined advertisers); pass other statuses, e.g. INVITE_PENDING or APPLICATION_PENDING, to see invitations and applications. sort is -partnership.statusUpdatedAt (newest first) unless given. Pass a publisher_id from awin_list_accounts. Use the advertiser ids with awin_get_advertiser_terms. Errors: account_not_accessible means the user cannot use this endpoint on the publisher (not linked, role too low or not entitled), never empty data; obo_exchange_refused and obo_exchange_failed mean the request could not be authorised. Report these and do not retry in this turn. If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_list_programmes

ChatGPT
List programmes for a publisher (GET /publishers/{id}/programmes). Pass a publisher_id from awin_list_accounts. relationship defaults to joined. Does not return date-range performance; use awin_get_advertiser_performance for clicks, sales, and commission. Docs: https://help.awin.com/apidocs/get-program-information If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_list_transaction_queries

ChatGPT
Lists the transaction queries publishers have raised against an advertiser account: claims for commission on sales that were not tracked (UNTRACKED), were tracked with the wrong amounts (INCORRECT) or were declined (DECLINED), each with its status (PENDING, APPROVED or DECLINED), order reference, claimed sale amount and commission. Use it to find queries waiting on the advertiser (status=PENDING) or to review past decisions. Pass an advertiser_id from awin_list_accounts. date_type picks the date start_date and end_date (yyyy-MM-dd, inclusive, UTC) apply to: transaction (when the sale took place), enquiry (when the query was raised) or validation (when it was approved or declined). The range is at most 31 days; a longer range returns date_range_exceeded without calling the API. Results are newest first on that date unless sort is given. duplicateTransactionId, when present, means a transaction already exists for the same order reference, so approving the query may pay for the sale twice; point this out. An empty list means no query matches, not an error. This tool does not approve or decline queries. When incomplete is true, not every query was read: narrow the range or filters and never report the list as complete. Errors: account_not_accessible means the user cannot use this endpoint on the advertiser (not linked, role too low or not entitled), never empty data; obo_exchange_refused and obo_exchange_failed mean the request could not be authorised. Report these and do not retry in this turn. If any list is larger than limit rows or the result exceeds the serialized size cap, the response is a bounded envelope (full-set summary on the first overflow call only, then truncated, has_more, next_cursor) instead of the raw payload. A single list is returned as items / total_count; several lists are named collections each with items, total_count, and has_more. limit applies to each list. A page is cut to fit the size cap: when the requested rows do not fit, fewer rows are returned (at least one per list), rows_truncated is true, and has_more / next_cursor continue from the rows actually returned. Pass limit 0 to return the complete raw payload only when it fits the size cap; a larger result returns error result_too_large with total_count instead of the data. Pass next_cursor with the same other arguments and the same limit to page every list, or pass offset (0-based, default null) with limit to jump every list to that slice. If both offset and cursor are passed, offset wins. Nested lists on a page are shortened to fit the size cap and are not pageable. When the call includes start_date and end_date, the result includes window with those requested dates, and each named collection does too. If the full result already fits, the payload is the complete raw response.

awin_search_help

ChatGPT
Search Awin's knowledge base for how-to answers (e.g. tracking setup, account configuration, platform usage). Returns grounded excerpts with source links. Does not return customer performance or report data. Always call awin_list_accounts before searching. Pass account_type when all accounts share one type, or when the user has a mixed portfolio and has explicitly said which account they are using. If accounts are mixed and the user has not said which account they are using, ask before calling awin_search_help. Do not infer account_type from topic words (e.g. asking about publishers does not mean the user is a publisher account).

give_feedback

ChatGPT
Report a problem with this MCP server to the Awin team: an unexpected error, a confusing or incomplete result, missing instructions, or a task these tools could not complete. The report is recorded for the team to review; it does not fix anything in this session and returns no data. Describe the goal, what you tried and where you got stuck. Do not include personal data, credentials or report figures. Does not call Awin APIs.

list_skills

ChatGPT
List the Awin skill guides. A skill explains how to combine these tools for a task, such as reviewing how an advertiser programme is doing or reporting on a tagged group of partners: which tools to call, in what order, and common pitfalls. Call this before load_skill whenever you do not already know the exact skill name; names are not predictable from topic words. Load skills proactively when starting a multi-step task, not after errors. Use query to search names and descriptions. Does not call Awin APIs.

load_skill

ChatGPT
Load an Awin skill guide before using the tools it covers, and follow it. If you do not already know the exact skill name from list_skills, call list_skills first; do not guess names. Set header_only=true to preview a skill's description and bundled resources. Set resource_path to load one bundled resource. Does not call Awin APIs.

whoami

ChatGPT
Return the authenticated principal from the Auth0 access token. Does not call Awin APIs.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/awin"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/awin)