First Meta review
Enable stored reporting for one Meta account, review its performance and retrieve the relevant creative context.
Choose one Meta ad account and keep the reporting period in that account's time zone. The hosted MCP and the local adcom mcp serve bridge expose the same tools.
Enable reporting
Call adcomputer_list_accounts and choose an explicit account_id. The result includes currency, time zone, sync progress and stored date coverage. Use refresh: true if account access needs revalidation.
Enable background ingestion for that account:
{
"account_id": "act_123",
"operation": { "action": "enable", "history_days": 90 }
}Send this to meta_sync. It returns a durable run ID immediately. Check adcomputer_list_accounts for progress. The first run publishes the latest seven calendar days, followed by older backfill.
Query performance
Call meta_describe with the account and reporting dates to discover metric definitions, units, supported object fields and raw event keys observed in that period. Select sections: ["actions"] when you only need conversion-event keys. An absent key is not proof that an event is unsupported.
Call meta_report with explicit dates. Replace these example dates with your chosen period:
{
"account_id": "act_123",
"time_range": { "since": "2026-10-10", "until": "2026-10-16" },
"level": "campaign",
"metrics": ["spend", "impressions", "clicks", "ctr", "cpc"],
"limit": 50
}The result includes whole-query totals, coverage and freshness. Missing partitions produce not_ready or incomplete; they do not imply zero activity. Today's data is provisional. A cursor keeps the same snapshot while you read additional pages.
Fictional response excerpt. This illustrates the request above, as if read at 2026-10-17T09:00:00Z. All values are invented. The snapshot and cursor are placeholders. Only two rows from a larger first page are shown. A full response also contains the remaining rows, report_spec, attribution and exact_metrics.
{
"schema_version": "meta_reporting_v1",
"state": "ready",
"source": "clickhouse",
"account_id": "act_123",
"currency": "USD",
"timezone": "America/New_York",
"time_range": { "since": "2026-10-10", "until": "2026-10-16" },
"level": "campaign",
"time_increment": "all_days",
"snapshot_id": "fictional-snapshot-id",
"rows": [
{
"object_id": "1001",
"name": "Example campaign A",
"spend": 100,
"impressions": 10000,
"clicks": 200,
"ctr": 2,
"cpc": 0.5
},
{
"object_id": "1002",
"name": "Example campaign B",
"spend": 50,
"impressions": 5000,
"clicks": 100,
"ctr": 2,
"cpc": 0.5
}
],
"totals": {
"spend": 200,
"impressions": 20000,
"clicks": 400,
"ctr": 2,
"cpc": 0.5
},
"missing_source_metrics": [],
"coverage": {
"complete": true,
"missing_days": [],
"provisional_days": [],
"stale": false,
"oldest_synced_at": "2026-10-17T08:00:00Z",
"newest_synced_at": "2026-10-17T08:05:00Z"
},
"paging": {
"has_more": true,
"after": "fictional-opaque-cursor"
}
}spendandcpcuse major currency units, USD here.clickscounts all ad clicks.ctr: 2means 2%, calculated from clicks and impressions.totalscovers the whole query, including omitted rows and later pages. The two displayed rows spend USD 150; the whole query spends USD 200. Total CTR and CPC are calculated from whole-query sums.coverage.completeandmissing_daysdescribe published dates.oldest_synced_atandnewest_synced_atshow the range of partition sync times. At the fictional read time,stale: falsemeans none exceed the freshness thresholds. No dates are provisional because the period excludes that account's current day.snapshot_ididentifies the published snapshot. Whenpaging.afteris present, reuse that opaque value unchanged asafterin the nextmeta_reportrequest with the same account, dates, filters, metrics and sort. Use the real response's value, not this placeholder. The cursor preserves the snapshot for one hour;paging.has_more: falsemarks the last page.
For conversions or ROAS, add the desired metrics and one explicit action_type, such as offsite_conversion.fb_pixel_purchase. The server never combines overlapping action types. Ratios are calculated from totals.
Use the IDs from the report with meta_objects to inspect campaigns, ad sets, ads or referenced creatives. Batch IDs with object_ids and select only needed fields, such as ["id", "name", "effective_status"] for ads. Missing IDs and unavailable fields are explicit. Those objects are last observed configuration, not configuration reconstructed at a historical reporting date.
The reporting tools declare versioned output schemas. Reports identify their exact snapshot and object responses identify currency and budget units. Pages can be smaller than limit to keep the complete tool result below 64 KiB; continue with paging.after. Add include_execution: true for server timing, query IDs and scanned rows/bytes.
Request a refresh or pause
Use meta_sync with operation.action set to refresh and an optional inclusive time_range. Repeated requests while a run is active return the current job and its actual range. Pause with operation.action set to pause; already published reports remain available.
Account sync status includes the active phase, planned waits, stale-refresh flags and the last technical reconciliation against Meta. Reconciliation checks data fidelity; marketing interpretation remains with you and your agent.
Exact live queries
Use meta_insights for reach, frequency, unique metrics, breakdowns, older dates and custom attribution windows. Daily unique counts cannot be added into an exact period total. You can also use this tool for an immediate review while the first sync is still running.
The reporting workflow changes local ingestion state. It does not change Meta campaigns, ads or budgets.