Setting Up A Data Warehouse Source

Use this skill when the user wants to connect an external data source to PostHog’s data warehouse for the first time.

Default to the one-step flow: data-warehouse-source-setup validates credentials, discovers every table, enables them with sensible sync defaults (incremental where possible), and creates the source in a single call — no schemas array to assemble. For credentials, hand the user a secure browser link with data-warehouse-source-connect-link instead of collecting secrets in chat. Only drop to the manual wizard → db-schema → create flow when the user wants to hand-pick which tables sync or set non-default sync types per table.

When to use this skill

• The user wants to connect a new source: “connect Stripe”, “import my Postgres orders table”, “sync Hubspot contacts”
• The user isn’t sure what source types PostHog supports
• The user has credentials but doesn’t know how to structure the schemas payload
• The user wants guidance on which sync method to pick per table

Available tools

Recommended: one-step setup

Most setups should use this path — it avoids the most common failures (skipping db-schema, malformed schemas, missing required fields).

1. Discover the source type and its fields (optional): external-data-sources-wizard lists every source type and the credential fields each needs. Use it to know what to ask the user for; skip it if the source type is obvious.

2. Collect credentials securely: call data-warehouse-source-connect-link({source_type}). It returns a connect_url to a minimal connect page rendering the source’s full connection form — the user authorizes via OAuth or enters credentials there, whichever the source offers (the response’s auth_method tells you which to expect). The page validates the details against a live connection and stashes them encrypted in a temporary store — it does NOT create the source. After the user confirms they’re done, find the stored credential id via data-warehouse-stored-credentials-list (filter by source_type, newest first; the page also shows the id to the user) and pass {"credential_id": <id>} to setup. Stored credentials are single-use — deleted as soon as setup consumes them — and expire after 24 hours.

   Never ask the user to paste raw database passwords, API keys, or OAuth tokens into the chat.

3. Create in one call: data-warehouse-source-setup({source_type, payload, prefix}). The server validates credentials, discovers all tables, enables them with sync defaults (incremental where a tracking column exists, else append, else full_refresh — never CDC), sets created_via=mcp, and creates the source. The payload carries a credential reference ({"credential_id": ...} or the OAuth integration id key) — or inline credentials for headless automation; no schemas array is needed. On success you get the new source id; call external-data-schemas-list to show the user what was enabled and how each table will sync.

Notes specific to this path:

• All discovered tables are enabled. That’s intended (incremental defaults keep ongoing cost low), but flag row counts for very large tables and offer the advanced flow if the user wants to sync only a subset.
• Webhooks are auto-registered for sources that support them (currently Stripe). Check the webhook key in the setup response: on success, webhook-capable tables sync in real time and webhook-only tables (e.g. Stripe Discount) are enabled too; on failure (e.g. the API key can’t create webhooks), tables keep the polling defaults and webhook-only tables stay disabled — relay the webhook.error to the user and offer Step 6 to register manually. If webhook.pending_inputs is non-empty, collect those values and submit via external-data-sources-update-webhook-inputs-create. CDC is never chosen automatically; use the advanced flow + CDC steps for near-real-time Postgres.
• Inline credentials in payload still work for headless/automation, but prefer the connect-link handoff above.

Advanced: hand-pick tables (three-step flow)

Use this when the user wants to choose exactly which tables sync or set non-default sync types. Don’t try to shortcut to external-data-sources-create — you need the db-schema response to build a valid schemas payload.

         ┌────────────────────┐
         │ 1. wizard          │  What source types exist? What fields does each need?
         └────────┬───────────┘
                  ▼
         ┌────────────────────┐
         │ 2. db-schema       │  Validate creds. List tables + available sync methods per table.
         └────────┬───────────┘
                  ▼
         ┌────────────────────┐
         │ 3. create          │  Send source_type + credentials + schemas[] to actually create.
         └────────────────────┘

Workflow

Step 1 — Discover the source type

Call external-data-sources-wizard (no params). The response is a dict keyed by source type. Each entry describes:

• name — the canonical source_type string you’ll pass to later calls (e.g. "Postgres", "Stripe", "Hubspot").
• label / caption — human-readable.
• fields — the config fields needed (host, port, database, api_key, client_id/secret, …). Each has name, type (input, password, switch, select, file-upload), and required.
• featured, unreleasedSource — use to gauge readiness. Skip sources marked unreleasedSource: true unless the user explicitly asked for a preview.

Match the user’s request to a source. If they said “Postgres”, look up Postgres. If they said something ambiguous like “database”, present the top relevant matches (Postgres, MySQL, MongoDB, BigQuery, Snowflake, Redshift) and let them pick.

For OAuth-based sources (Hubspot, Salesforce, Google Ads), the wizard entry hints at an OAuth flow. These typically need the user to authorize in the PostHog UI rather than pasting credentials — explain this and direct them to the source setup page rather than trying to collect tokens in chat. OAuth is about authentication, not about how data flows; OAuth sources still use polling bulk sync, not webhooks.

Gather the required credentials from the user. Never ask for more fields than the wizard entry says are required — asking for an unnecessary port when the source doesn’t need one confuses users.

Step 2 — Validate credentials and discover tables

Call external-data-sources-db-schema with source_type plus all credential fields. This does two things at once:

1. Validates the credentials against the live source. Returns 400 with a message if anything is wrong (bad host, wrong password, permission denied). Show the error verbatim — it’s often actionable (“password authentication failed for user ‘x’”).
2. If valid, returns an array of table entries. Each entry:

{
  "table": "orders",
  "should_sync": false,
  "rows": 1_250_000,
  "incremental_available": true,   # can do sync_type=incremental
  "append_available": true,        # can do sync_type=append
  "cdc_available": true,           # can do sync_type=cdc  (null = not enabled for team)
  "supports_webhooks": false,      # can do sync_type=webhook for real-time push
  "incremental_fields": [          # candidates: usually updated_at, created_at, id
    {"field": "updated_at", "type": "datetime", "label": "updated_at", ...},
    {"field": "created_at", "type": "datetime", ...},
    {"field": "id", "type": "integer", ...}
  ],
  "detected_primary_keys": ["id"],
  "available_columns": [{"field": "id", "type": "integer", "nullable": false}, ...],
  "description": "..."
}

Present this to the user. Don’t dump the raw JSON — summarize: which tables were found, row counts, and the default sync method recommendation per table (see sync-type decision guide (opens in a new tab)).

Step 3 — Confirm per-table sync configuration

For each table the user wants to sync, pick a sync_type. See the sync-type decision guide (opens in a new tab) for detailed rules, but the short version is:

• Small / dimension tables (<50k rows, no natural ordering column): full_refresh — simple and always correct.
• Large tables with an updated_at / modified_at: incremental — much cheaper per sync.
• Append-only immutable tables (logs, events): append if available — preserves history.
• Postgres with CDC enabled and you need near-real-time: cdc — requires primary keys and Postgres prerequisites.
• Sources that support webhooks (currently Stripe): for near-real-time ingestion set sync_type: "webhook" on the tables where supports_webhooks: true, then register the webhook as a post-create step (see step 6 below). Tables that don’t support webhooks on the same source still need a bulk sync_type.

For each schema that will use incremental/append/cdc, you also need:

• incremental_field — which column to track for high-water-mark ordering. Pick from the incremental_fields list returned by db-schema. Prefer updated_at over created_at (updated_at catches late-arriving updates; created_at misses them). For integer-only tables, use the monotonically increasing primary key.
• incremental_field_type — must match the chosen field’s type (datetime, timestamp, date, integer, numeric, objectid).
• primary_key_columns — required for CDC. Use detected_primary_keys from db-schema.

Step 4 — Pick a good prefix

The source’s prefix is prepended to table names in HogQL. Tables end up as {prefix}_{table_name}.

• Default to the source type lowercased if there’s only one source of that type: stripe, postgres.
• If the user already has a Postgres source, pick something distinguishing: postgres_prod, postgres_analytics.
• Use lowercase, underscore-separated. The prefix becomes part of every HogQL query the user writes.

Confirm the prefix with the user before creating — changing it later is possible but renames every table.

Step 5 — Create the source

Call external-data-sources-create with:

{
  "source_type": "Postgres",
  "prefix": "postgres_prod",
  "payload": {
    "host": "...",
    "port": "5432",
    "dbname": "...",
    "user": "...",
    "password": 

Rules for the schemas array:

• Every table returned by db-schema should be included, even ones the user doesn’t want (set should_sync: false). Tables the user didn’t mention default to should_sync: false.
• sync_type is required only when should_sync: true.
• incremental_field / incremental_field_type must be present when sync_type is incremental or append.
• primary_key_columns must be present when sync_type is cdc.

On success you’ll get back a source with a new id. The first sync is triggered automatically.

Step 6 — Register a webhook (only when any schema is sync_type: "webhook")

Webhook-type schemas don’t start receiving data just by existing — the external service needs to know where to POST events, and PostHog needs to know how to verify them. This is a second call after source creation, not part of the external-data-sources-create payload. Do this before telling the user the setup is complete, otherwise they hear “syncs are running” while the push channel is still unregistered.

Only needed when at least one schema on the source has sync_type: "webhook" and should_sync: true. Currently only Stripe implements this flow; for everything else skip this step.

Before calling create-webhook, check external-data-sources-webhook-info-retrieve({id}). If it already returns exists: true, do NOT call create-webhook again — each successful call registers a new external endpoint and would result in duplicate deliveries.

1. Call external-data-sources-create-webhook-create({id}). PostHog:

   • creates the HogFunction that will receive webhook POSTs,
   • builds a schema_mapping from external event types to PostHog schema ids,
   • calls the source’s API (e.g. Stripe) to register the webhook URL and subscribe to the relevant events,
   • on Stripe, auto-captures the signing_secret and stores it securely.

   Returns {success, webhook_url, error}. On success report the webhook_url to the user for their records — but they don’t need to paste it anywhere; registration is already done.

2. If success: false with a permissions error like “API key doesn’t have permission to create webhooks”:

   • The HogFunction is still created, just disabled.
   • Ask the user to create the webhook manually in the source’s dashboard using the returned webhook_url.
   • Have them copy the signing secret from the source’s webhook settings.
   • Call external-data-sources-update-webhook-inputs-create({id}, {inputs: {signing_secret: "whsec_..."}}) to store it. The HogFunction picks it up and verifies incoming payloads.

3. Verify with external-data-sources-webhook-info-retrieve({id}). A healthy webhook has exists: true, external_status.status: "enabled", and no error.

Webhooks are supplementary to bulk sync. The first load of a webhook-enabled schema is still done via polling (initial_sync_complete flips to true when done); after that, the webhook becomes the primary ingestion path. A webhook schema will still have a sync_frequency that schedules a periodic bulk refresh as a safety net. This is expected — not something to “fix”.

Step 7 — Confirm and explain what happens next

After creation (and, for webhook schemas, after Step 6):

• Call external-data-schemas-list to show the user the initial state.
• Explain: every enabled schema enters Running, then moves to Completed when the first sync finishes. First syncs can take anywhere from seconds to hours depending on row count — a multi-million-row table is fine, just slow.
• Tell them how to query: SELECT * FROM {prefix}_{table_name} LIMIT 10 in HogQL.
• Offer to check back in a few minutes to confirm the initial syncs succeeded.

CDC setup for Postgres (optional, when requested)

If the user wants near-real-time replication from Postgres:

1. Before calling db-schema, run external-data-sources-check-cdc-prerequisites-create with their Postgres creds. It returns {valid, errors[]} listing anything missing (wal_level, replication slot, publication, permissions).
2. If valid: false, present the errors and ask the user to fix on the Postgres side. Don’t try to create a CDC source that will immediately fail.
3. Once prerequisites pass, proceed to db-schema and create. Set sync_type: "cdc" on the tables that need it, and include primary_key_columns for each (CDC requires them).

Important notes

• Always validate creds with db-schema before create. The create endpoint will accept invalid creds and then fail asynchronously — the source appears in the list with status Error and no tables. Skipping the validation step just pushes the failure into the background.
• Present the table list before creating. Large databases may have hundreds of tables. Don’t auto-select them all — row counts and relevance matter for billing. Let the user opt in explicitly.
• Don’t invent schemas. Every entry in the schemas array must correspond to a real table from the db-schema response. You can’t “also add an orders table” unless db-schema found one.
• Prefix is load-bearing. It’s part of every HogQL query the user will ever write against these tables. Pick something short, descriptive, and not already taken.
• Prefer the secure connect-link for any credentials. Use data-warehouse-source-connect-link so the user authenticates in their browser — the connect page renders the source’s full connection form (OAuth and credential options alike) and stores the result without creating the source. Don’t collect OAuth tokens or database passwords in chat; pass the credential_id reference to setup — source creation always happens through setup, not the UI. (An already-connected OAuth integration can also be passed directly via its id key, e.g. {"hubspot_integration_id": 123}.)
• Webhooks are a separate step after create. Setting sync_type: "webhook" on a schema doesn’t register the webhook — the create-webhook call does. Always follow create → create-webhook → webhook-info for webhook-type schemas, and never leave a webhook schema dangling without registration (it just won’t receive events).
• Webhook support is source-specific and sparse. Currently only Stripe implements WebhookSource. Don’t promise webhooks for Hubspot, Salesforce, or Postgres — they’ll use polling sync.
• Row counts drive billing. Warehouse syncing is metered by rows synced. A chatty 500M-row events table synced hourly is very different from a 10k-row dimension table synced daily. Flag large tables and offer longer sync frequencies (sync_frequency: "24hour") as the default.