`aidp-connectors-overview` — pick the right connector skill

aidp-connectors-overview — pick the right connector skill

When to use

• The user is exploring options ("how do I connect to Oracle from AIDP?", "which connector should I use?").
• The user mentions multiple Oracle sources at once.
• The user describes a source by capability (e.g. "OLAP cube", "structured streaming") rather than naming the product directly.

When NOT to use

• The user has already named their target product (ALH, ATP, Fusion, etc.) — invoke the matching aidp-<product> skill directly.

How to route

Before any connector skill works, the helper package must be uploaded to the user's AIDP workspace. If the user hasn't done this yet (or you see ModuleNotFoundError: oracle_ai_data_platform_connectors in any prior cell), invoke aidp-connectors-bootstrap first. It uses the AIDP MCP tools to push the package into /Workspace/Shared/ and runs a sanity-import notebook.

Otherwise, pick the right skill from this table and invoke that skill. Don't re-write its content here.

Oracle / OCI sources

User says...	Use skill
"ALH", "AI Lakehouse", "ADW", "ATP", "Autonomous Database", "26ai"	aidp-alh — Autonomous DB family (wallet / IAM DB-Token / API key)
"Oracle Database", "generic Oracle DB", "on-prem Oracle", "Oracle 19c / 21c", "Base DB", "Oracle on Compute", non-Autonomous Oracle	aidp-oracle-db — generic Oracle DB via plain user/password on TCP 1521
"ExaCS", "Exadata", "Exadata Cloud", "private-subnet Oracle DB"	aidp-exacs
"PeopleSoft", "PSFT", "PS HCM", "FSCM", "Campus Solutions"	aidp-peoplesoft
"Siebel", "Siebel CRM", "S_CONTACT", "S_ORG_EXT"	aidp-siebel
"Fusion ERP", "Fusion HCM", "Fusion REST", "FA REST", "Cloud ERP API"	aidp-fusion-rest
"BICC", "BI Cloud Connector", "Fusion bulk extract", >50k rows from Fusion	aidp-fusion-bicc
"EPM Cloud", "EPBCS", "Hyperion Planning", "Planning app", "exportdataslice"	aidp-epm-cloud
"Essbase", "Essbase 21c", "MDX", "OLAP cube", "cube REST"	aidp-essbase
"OCI Streaming", "Kafka on OCI", "stream pool", "structured streaming Kafka"	aidp-streaming-kafka
"OCI Object Storage", "oci://", "external volume", "external table on bucket"	aidp-object-storage
"Iceberg", "Apache Iceberg", "time travel", "snapshots", "schema evolution"	aidp-iceberg

External RDBMS / Hadoop (non-Oracle)

User says...	Use skill
"PostgreSQL", "Postgres", "psql"	aidp-postgresql
"MySQL", "HeatWave", "MDS", "MySQL Database Service"	aidp-mysql
"SQL Server", "MSSQL", "Azure SQL", "TDS"	aidp-sqlserver
"Hive", "HiveServer2", "HS2", "HCatalog", non-Kerberos Hive	aidp-hive

SaaS

User says...	Use skill
"Salesforce", "SFDC", "Sales Cloud", "Service Cloud", "sObject", "SOQL", "Account / Opportunity / Lead"	aidp-salesforce

Multi-cloud + escape hatches

User says...	Use skill
"Snowflake", "sfUrl", "sfWarehouse"	aidp-snowflake
"ADLS", "Azure Data Lake", "abfss"	aidp-azure-adls
"S3", "AWS S3", "s3a"	aidp-aws-s3
"Generic REST", "manifest URL", "manifest.path", REST endpoint with manifest schema	aidp-rest-generic
"Custom JDBC", "ClickHouse", "DuckDB", "DB2", "SAP HANA", any DB without a dedicated skill	aidp-jdbc-custom
".xlsx", "Excel", "spreadsheet ingestion"	aidp-excel

What's blocked at the AIDP platform level (so you don't try)

• Instance Principal — IMDS (169.254.169.254) is unreachable from AIDP notebooks; signer either fails or runs in AIDP's service tenancy, not the customer's.
• Resource Principal — AIDP sets AIDP_AUTH=resource_principal but does NOT provide OCI_RESOURCE_PRINCIPAL_RPST / OCI_RESOURCE_PRINCIPAL_PRIVATE_PEM, so get_resource_principals_signer() raises.

If the user wants either of those, point them at API Key + inline OCI config (oracle_ai_data_platform_connectors.auth.from_inline_pem) instead. The AIDP team is aware; pending Oracle action.

Cross-cutting AIDP gotchas (every connector inherits these)

1. Credentials live under /tmp/ — never /Workspace/. The latter is FUSE-mounted; intermittent disconnects + os.chmod no-op.
2. Files written for the JDBC driver process must be world-readable up-front via os.open(..., O_WRONLY|O_CREAT, 0o666).
3. Spark streaming checkpoints must live under /Volumes/<catalog>/<schema>/<volume>/..., never /Workspace/, never oci://.
4. Refresh the AIDP session token before live testing: oci session authenticate --profile AIDP_SESSION --region us-ashburn-1.

References

• Plugin README: ../../README.md
• Live-test matrix + results: ../../tests/live-results/RESULTS.md
• AIDP notebook auth investigation: Claude context/AIDP/AIDP Context/AIDP/aidp-notebook-authentication.md