How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

platform-salesforce-connect-adapter-generate - Skill

Imported from forcedotcom/sf-skills (skills/platform-salesforce-connect-adapter-generate/SKILL.md). Install upstream with npx skills add forcedotcom/sf-skills --skill platform-salesforce-connect-adapter-generate. Copyright stays with the author.

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?

The user wants to…	Right tool
Query external data live without copying it — read-only or read-write, appears as Salesforce records	This skill
Copy or sync data into Salesforce on a schedule	the platform-data-manage skill
Connect to OData, GraphQL, DynamoDB, Athena, or Cross-Org	Standard adapter setup — no Apex needed, different flow
Connect to Snowflake via Salesforce's native Snowflake adapter (direct Snowflake protocol)	Built-in Snowflake adapter — no Apex needed
Access Snowflake (or any database) data via a REST or HTTP API	This skill — Snowflake REST endpoint → custom Apex adapter
Call an external API from Apex or Flow logic	the platform-apex-generate skill or integration-connectivity-generate
Expose Salesforce data to an external system	integration-connectivity-generate
Receive real-time pushed data or subscribe to external events	Not Salesforce Connect — Connect is pull-only. Use Platform Events or Change Data Capture instead
Sync or copy data for analytics or bulk processing	Data Cloud or ETL — Connect is zero-copy virtualization only

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:

Input	Why it matters
API name and what it does	Names the classes and sets context for field mapping
Base URL	Becomes the Named Credential endpoint
Auth type	Determines Named Credential setup and `getAuthenticationMode()`
Key GET endpoint(s) + sample response	Defines the External Object schema — field names, types, nesting
Write support needed?	Determines whether to implement `upsertRows()` and `deleteRows()`
Org type (DE, scratch, sandbox)	Sets `apiVersion` in `.cls-meta.xml`

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

External API type	Use this
string / text	`DataSource.DataType.TEXT_TYPE`
number / integer / decimal	`DataSource.DataType.NUMBER_TYPE`
boolean	`DataSource.DataType.BOOLEAN_TYPE`
date (ISO 8601)	`DataSource.DataType.DATE_TYPE`
datetime / timestamp	`DataSource.DataType.DATETIME_TYPE`
URL	`DataSource.DataType.URL_TYPE`
email	`DataSource.DataType.EMAIL_TYPE`
phone	`DataSource.DataType.PHONE_TYPE`
enum / picklist	`TEXT_TYPE` — map enum values as strings
nested object / JSON blob	Flatten to scalar fields, or `TEXT_TYPE` and parse in `query()`
array	Derive a count field (`NUMBER_TYPE`), or flatten first-element fields
string > 255 chars	Long text area — do not truncate; Salesforce maps it automatically

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:

Scenario	Auth mode	Extra capabilities	Named Credential
Public read-only	`ANONYMOUS`	`QUERY`, `SEARCH`	Deploy as metadata
Authenticated	`NAMED_PRINCIPAL` or `PER_USER`	`QUERY`, `SEARCH`	Create manually in Setup — never deploy credentials as metadata
Read-write	`ANONYMOUS` or auth	Add `ROW_CREATE`, `ROW_UPDATE`, `ROW_DELETE`	Per above
Paginated	Any	Any	Use `context.tableSelection.numberOfRows` (default 500 if null); no automatic `queryMore`

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:

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

Troubleshooting: 0 records with no error

Symptom	Cause	Fix
SOQL returns 0	Remote Site Setting missing	Setup → Remote Site Settings → New
SOQL returns 0	Named Credential not found	Setup → Named Credentials → confirm `callout:` name matches exactly
SOQL returns 0	Org proxy blocks external URLs	Use loopback pattern or switch to an external DE org
List view shows 0	Deployment Status = In Development	Object Manager → [Object]__x → Edit → Deployed
List view shows 0	Filter is "Recently Viewed"	Change filter to "All"
App Launcher shows nothing	No tab created	Setup → Tabs → Custom Object Tabs → New

What the agent deploys vs. what the developer does manually

Agent deploys (one `sf project deploy start` command)

File	Notes
`classes/[API]DataSourceConnection.cls` + meta	Always
`classes/[API]DataSourceProvider.cls` + meta	Always
`namedCredentials/[API].namedCredential-meta.xml`	Public APIs only — skip for any API requiring credentials

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.

Situation	Handle it by
HTTP 401	`throw new DataSource.OAuthTokenExpiredException()`
HTTP 429 / rate limit	Return empty rows; log via `System.debug()`
HTTP 500 or network error	Return empty rows; do not rethrow
Null field in response	`record.get('field') != null ? (String) record.get('field') : ''`
Nested JSON that won't flatten	Return raw JSON string as a TEXT_TYPE field rather than failing the whole query

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:

Deploy Apex classes first (sf project deploy start)
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

Don't	Why	Do
Generate code without knowing the API shape	Wrong field types, wrong ExternalId mapping, broken adapter	Ask for API name, base URL, auth type, sample response first
Deploy ExternalDataSource as metadata	Not a deployable metadata type — sf CLI will error	Register the External Data Source manually in Setup after deploying the Apex classes
Use `upsertRow()` or `deleteRow()` (singular)	These methods don't exist — compile error	Use `upsertRows()` and `deleteRows()` (plural)
Put `sync()` only on Provider	On some API versions `sync()` must be on Connection — compiler will tell you	Implement `sync()` on Connection; remove from Provider if it errors
Hardcode the API URL in Apex	Credentials exposed in source, callout blocked	Always use `callout:NamedCredentialName`
Return a single `DataSource.TableResult` from `search()`	`search()` returns `List<DataSource.TableResult>` — compile error	Loop over `context.tableSelections` and return one result per table
Skip COUNT handling in `query()`	List views fire COUNT queries; without detection they return wrong results	Check `columnsSelected[0].aggregation == DataSource.QueryAggregation.COUNT`
Leave External Object in "In Development"	Records return 0 with no error — invisible to the user	After Validate and Sync: Object Manager → [Object]__x → Edit → Deployed
Skip the Remote Site Setting	Callout fires silently, returns empty rows, no exception	Deploy `remoteSiteSettings/[API].remoteSite-meta.xml` or add manually in Setup

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:

force-app/main/default/classes/[API]DataSourceConnection.cls
force-app/main/default/classes/[API]DataSourceProvider.cls
force-app/main/default/classes/[API]DataSourceConnection.cls-meta.xml
force-app/main/default/classes/[API]DataSourceProvider.cls-meta.xml
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.