All docs
Connect

Adobe Experience Platform: connect a sandbox, permissions and tools

TL;DR

Adobe Experience Platform connects on its own Integrations card with an OAuth Server-to-Server credential, your IMS Organization ID and one sandbox; the credential's product profile decides which schemas, datasets and audiences the agent can read or change.

Before you start (in Adobe)

  1. 1Open Adobe Developer Console (developer.adobe.com/console), pick your organization, and open a project (or Create new project).
  2. 2Add API → Experience Platform API, choose OAuth Server-to-Server, and tick the product profiles the agent should use.
  3. 3In Adobe Admin Console → Adobe Experience Platform → the profile, grant the sandboxes and permissions below.
  4. 4Copy the Organization ID (ends in @AdobeOrg) with the Client ID and secret, or press Download for Project JSON.

Connect

  1. 1Integrations → Adobe Experience Platform → Connect → Paste JSON (or Manual entry: name, Client ID, secret, IMS Organization ID).
  2. 2Test connection lists the sandboxes and reads one schema-registry class in the chosen sandbox.
  3. 3Pick the sandbox (prod or a development sandbox) and Save connection. A call can name another sandbox the credential reaches.
Scopes are added for you: openid, AdobeID, read_organizations, additional_info.projectedProductContext, session.

Which permissions unlock what

  • Sandboxes: which sandboxes the credential reaches.
  • Data Modeling: schemas and field groups (read and write).
  • Data Management: datasets (403 without it).
  • Profile Management: audiences and merge policies (401 "RBAC Unauthorized" without it).
  • Identity Management: identity namespaces. Query Service: queries (403 without it).

Every Adobe Experience Platform tool

5 reads and 2 writes, plus get_connection and test_connection shared by the four Adobe products.

ToolKindWhat it does
aep_list_sandboxesReadThe Experience Platform sandboxes this credential reaches (GET data/foundation/sandbox-management/): name, title, type (production | development), default, region.
aep_list_schemasReadXDM schemas in the connection's sandbox (GET data/foundation/schemaregistry/tenant/schemas, or global with container:"global"): $id, title, meta:class, version.
aep_list_datasetsReadDatasets in the sandbox (GET data/foundation/catalog/dataSets): id, name, schemaRef, tags, created, last batch.
aep_list_audiencesReadAudiences (segment definitions) in the sandbox (GET data/core/ups/segment/definitions): id, name, expression (PQL), evaluation (batch | streaming | edge), merge policy, status.
aep_getReadANY Experience Platform API GET (path starting data/: data/foundation/catalog/…, data/foundation/schemaregistry/…, data/core/ups/…, data/foundation/flowservice/…, data/foundation/identity/…).
aep_create_audienceWriteCreate an audience (segment definition) in the sandbox (POST data/core/ups/segment/definitions): { accountId?, name, description?, expression (PQL — e.g. homeAddress.country = "US"), mergePolicyId?, evaluation? (batch…
aep_mutateWriteAny other Experience Platform write (method POST | PUT | PATCH | DELETE, path data/…, body = Adobe's own fields): schemas, field groups, datasets, audiences, identity namespaces, tags. ★ DELETE, data ingestion (data/f…
Over MCP and the REST API the names are adobe__<tool> (for example adobe__aep_list_schemas). A workspace with no connection for this product does not list these tools over MCP; describe_tool names the card to connect. Setup for Claude, Codex or Cursor: MCP server guide.

Limits and formats

  • Schemas: up to 300 per call (tenant container by default, container:"global" for Adobe's). Datasets and audiences: up to 100 per call.
  • Schema Registry reads get the Accept header each path wants (xed-id for a list, xed for one resource); pass accept for another form, such as application/vnd.adobe.xed-full+json.
  • Every request carries the IMS organization and the sandbox; the sandbox list needs its trailing slash (set for you).

Changes and approvals

  • Ordinary changes follow the approval setting — by default each waits for your approval: an audience (aep_create_audience), a schema, a field group, an identity namespace, dataset tags.
  • A person approves, every time: data ingestion (data/foundation/import), dataflows and destinations (data/foundation/flowservice — activating an audience is one), merge policies, Profile system jobs, Query Service statements and any DELETE.
  • Refused on every lane: sandboxes, access control, privacy and data-hygiene jobs, the Catalog's multi-request root, and a body that carries a batch of operations.

Manage the connection

  • Test again re-checks this product with one read-only request and stores the result.
  • Configure changes this product's details; Replace credential swaps a rotated secret in place (cached tokens are dropped).
  • Disconnect deletes the stored credential and its tokens here. It does not revoke the credential in Adobe — do that in Developer Console.
More on Adobe Experience Platform: what the agent does · setup and every tool · MCP server · AI agent guide · playbook. The other Adobe products, each its own connection: Adobe Analytics · Adobe Target · Adobe Launch · All four together: Adobe.

Frequently asked questions

Why do datasets answer missing permissions?
The credential's Experience Platform product profile lacks Data Management. Add it in Adobe Admin Console and press Test connection.
Can the agent work in several sandboxes?
The connection has one default sandbox; a call can name another sandbox the credential reaches.
Does the agent ingest data?
Only with a person's approval, every time — ingestion, dataflows and destinations move data in or out of Experience Platform.

Related