Salesforce Connect — Custom Apex Adapter

Route the user through building a complete custom Salesforce Connect Apex adapter: two Apex classes, metadata deployment, and registration. Do not guess at the user's API shape — ask before generating.

Scope

In scope: Generating DataSource.Connection and DataSource.Provider Apex classes, Named Credential metadata, deployment via sf CLI, and step-by-step Setup registration guidance for any REST API.

Out of scope: Configuring standard adapters (OData, Snowflake, GraphQL — those have their own flows). Generating the External Data Source metadata file (not deployable via sf CLI — must be registered manually in Setup). Writing Apex that calls external APIs outside the Salesforce Connect framework (use the platform-apex-generate or integration-connectivity-generate skills).

Before starting

Confirm two things. If either is missing, ask before proceeding.

1. Is this actually a Salesforce Connect use case?

If the user has a standard adapter available, tell them — writing a custom adapter when a standard one fits is unnecessary work.

2. Do they have a Salesforce Connect license?

A custom adapter requires a Salesforce Connect license (one license per External Data Source). Without it, the External Data Source menu won't show the Apex option. If the user doesn't have one, tell them before going further.

Default org context (when running locally)

If the user does not specify an org alias, use demo-org. If the user does not specify an SFDX project path, use ~/salesforce-connect-apex-skill/sfconnect-demo/. These are the defaults for local testing — always override if the user specifies their own.

Collect inputs before generating

Never generate code without these. If any are missing, ask:

If the user provides an OpenAPI spec, extract:

• All GET endpoints returning arrays → each becomes a DataSource.Table
• Response object properties → DataSource.Column entries
• id / uuid / primary key → map to ExternalId
• Property types → use the field type mapping table below

What to build

Every custom Salesforce Connect adapter is exactly two Apex classes.

DataSource.Connection

Handles communication with the external API. Exact signatures:

override global DataSource.TableResult query(DataSource.QueryContext context)
override global List<DataSource.TableResult> search(DataSource.SearchContext context)

// Write support — only if API supports it:
global override List<DataSource.UpsertResult> upsertRows(DataSource.UpsertContext context)
global override List<DataSource.DeleteResult> deleteRows(DataSource.DeleteContext context)

Key distinction: query() always operates on one table (QueryContext has a single TableSelection). search() can operate on multiple tables simultaneously (SearchContext has multiple TableSelection instances) — handle each in a loop and return a result per table.

DataSource.Provider

Declares the adapter's capabilities and schema to Salesforce. Required methods:

• getAuthenticationMode() — return ANONYMOUS for public APIs; NAMED_PRINCIPAL or PER_USER for authenticated APIs
• getCapabilities() — declare QUERY, SEARCH; add ROW_CREATE, ROW_UPDATE, ROW_DELETE only if implementing writes
• getConnection(ConnectionParams params) — return new YourConnection(params)
• sync() — called when the user clicks "Validate and Sync" in Setup; returns List<DataSource.Table> defining the External Object schema and columns

The Provider class appears in Salesforce Setup as Custom-[ClassName] under the External Data Source type dropdown.

Critical: whenever you edit the Connection class, you must resave the Provider class too — even with no changes. Otherwise the adapter disappears from the Type picklist and existing External Object tabs break.

Field type mapping

Every DataSource.Table must include these columns:

DataSource.Column.text('ExternalId', 255)  // unique key from external system — REQUIRED
DataSource.Column.text('Name', 255)        // display label shown in Salesforce UI — REQUIRED
DataSource.Column.url('DisplayUrl')        // link to the record in the external system — recommended

Missing ExternalId is the most common reason an adapter deploys but records don't appear. DisplayUrl enables the clickable link icon in list views — populate it with the external record's URL.

Scenarios

Four scenarios: (1) public read-only, (2) authenticated API key/OAuth, (3) read-write with upsert/delete, (4) paginated API. Full code patterns for each are in references/scenarios.md.

Key differences by scenario:

How to verify the adapter is working

Option 1 — SOQL via sf CLI (fastest, no UI needed)

sf data query \
  --query "SELECT ExternalId, Name__c FROM [YourObject]__x LIMIT 5" \
  --target-org [org-alias]

If rows come back, the adapter is working. If 0 rows, check the troubleshooting steps below.

Option 2 — Visualforce page (best for demos, works on any org)

Deploy a Visualforce page + Apex controller that queries [Object]__x and renders rows in an apex:pageBlockTable. Open at [orgUrl]/apex/[PageName]. More reliable than list views for testing — bypasses tab requirements, deployment status, and filter issues. See references/scenarios.md for the standard template.

Option 3 — List view (standard UI, but requires extra steps)

For the list view to show records, all four of these must be true:

1. External Object Deployment Status = Deployed (Object Manager → Edit)
2. A tab exists for the External Object (Setup → Tabs → Custom Object Tabs → New)
3. List view filter is set to All (not "Recently Viewed")
4. Remote Site Setting exists for the external API URL

Troubleshooting: 0 records with no error

What the agent deploys vs. what the developer does manually

Agent deploys (one sf project deploy start command)

Do NOT generate externalDataSources/[API].externalDataSource-meta.xml. The External Data Source must be registered manually in Setup after deploying the Apex classes. The sf project deploy start command cannot deploy custom Apex External Data Sources — the sf CLI will error with "Could not infer a metadata type." Tell the user to go to Setup → External Data Sources → New after deployment.

Correct NamedCredential XML format (no <name> element — the API name comes from the filename):

<?xml version="1.0" encoding="UTF-8"?>
<NamedCredential xmlns="http://soap.sforce.com/2006/04/metadata">
    <label>My API</label>
    <endpoint>https://api.example.com</endpoint>
    <allowMergeFieldsInBody>false</allowMergeFieldsInBody>
    <allowMergeFieldsInHeader>false</allowMergeFieldsInHeader>
    <generateAuthorizationHeader>false</generateAuthorizationHeader>
    <principalType>Anonymous</principalType>
    <protocol>NoAuthentication</protocol>
</NamedCredential>

Deploy command:

sf project deploy start \
  --source-dir force-app/main/default/ \
  --target-org [org-alias]

Developer does manually (always — no metadata equivalent exists)

Validate and Sync triggers a live call to the external API to discover its schema and create External Object definitions. This cannot be scripted.

Setup → External Data Sources → [your data source] → Validate and Sync
Check the box next to each External Object → Sync
App Launcher → search [External Object name] → confirm records appear

Always end with: "Deployment complete. One step left: open the External Data Source in Salesforce Setup and click Validate and Sync."

Error handling rules

Never let an exception propagate uncaught out of query() — Salesforce shows a generic error to the user with no context.

Governor limits

• Max 100 HTTP callouts per transaction — each query() call is one transaction
• Heap limit 6 MB (synchronous) — don't deserialize massive payloads; use server-side pagination
• CPU limit 10 seconds — avoid nested loops over large response arrays

Setup prerequisites — in this exact order

1. Deploy Apex classes first

The Salesforce Connect: Custom (Developed with Apex) type only appears in the External Data Source type dropdown after at least one DataSource.Provider subclass is deployed in the org.

This is not a permissions issue. It is a dependency — Salesforce discovers available Provider classes at runtime. No deployed Provider = no option in the dropdown.

Correct sequence — always:

1. Deploy Apex classes first (sf project deploy start)
2. Then go to Setup → External Data Sources → New → the Custom Apex type will appear

If the user says "I don't see the Custom Apex option in the dropdown", the answer is: deploy the classes first, then come back to Setup.

2. Add Remote Site Setting before testing

The external API's base URL must be whitelisted or callouts fail silently — query() returns empty rows with no exception, no error message. Deploy as metadata or add at Setup → Remote Site Settings → New.

3. After Validate and Sync — set Deployment Status to Deployed

After Validate and Sync, Salesforce creates the External Object with status "In Development". Records return 0 results in list views and SOQL until this is changed. No error is shown — another silent failure.

Setup → Object Manager → [YourObject]__x → Edit
  Deployment Status → Deployed → Save

Always tell the user this step after Validate and Sync completes.

4. Create a tab to surface the External Object in App Launcher

External Objects are invisible in the UI without a tab. Create one at:

Setup → Tabs → Custom Object Tabs → New → select [YourObject]__x → Save

Also required

• Salesforce Connect license — without it, all Connect adapter types including OData are hidden. If the user sees OData and Cross-Org in the dropdown, the license is already active.

Anti-patterns

Before shipping

Confirm: inputs collected (API name, URL, auth, sample response, write support) → code has sync() + query() + search() on Connection, getCapabilities() + getConnection() on Provider, ExternalId and nameColumn on every table, COUNT handling in query(), mock data for tests, callout: prefix on all endpoints → both .cls-meta.xml files generated → deploy command given → post-deploy steps communicated (Named Credential setup, Remote Site Setting, External Data Source registered, Validate and Sync, Deployment Status → Deployed, tab created).

Reference examples

See references/official-examples.md for annotated study of the official Salesforce Connect adapter examples — GitHub Issues (full DML, picklist, cross-table relationships), Google Drive (OAuth + test mock), Google Books (pagination), StackOverflow (multiple tables), and Loopback (filter translation). Read before generating for a new user.

Project setup — if the developer has no SFDX project yet

Before deploying, they need a project structure. If one doesn't exist, generate it first:

sf project generate --name my-adapter --output-dir .
cd my-adapter

This creates force-app/main/default/classes/ and the required sfdx-project.json. All generated files go into this structure. The developer does not need VS Code, Agentforce Vibes, or any Salesforce IDE — just sf CLI installed and an org authenticated via sf org login web --alias my-org.

This skill works with any coding agent that can read a context file — Claude Code, Cursor, Windsurf, GitHub Copilot, or any agent with the SKILL.md loaded. No Salesforce MCP server, no Agentforce Vibes, no internal Salesforce tooling required.

Output format

Generate files in this order, then give the deploy command:

1. force-app/main/default/classes/[API]DataSourceConnection.cls
2. force-app/main/default/classes/[API]DataSourceProvider.cls
3. force-app/main/default/classes/[API]DataSourceConnection.cls-meta.xml
4. force-app/main/default/classes/[API]DataSourceProvider.cls-meta.xml
5. force-app/main/default/namedCredentials/[API].namedCredential-meta.xml — public APIs only

One-line comment at top of each Apex class: // Salesforce Connect custom adapter for [API name] No other inline comments unless a logic choice is non-obvious.