You ask for a review of all rules in your workspace. The AI reads each rule’s conditions, actions, schedule, and filters, then flags issues: overlapping conditions, contradictory actions (one pauses what another starts), thresholds with no volume guard, long-disabled rules taking up space.

Birch AI does this in-app with zero setup and can tie findings to live performance data. Birch MCP does this in Claude using list_automation_rules → get_rule_details per rule, and handles larger accounts (200+ rules tested) without timeouts.

Neither tool fixes rules for you. Both describe issues; you decide what to change.

You want to know what your rules actually did: which ones fired, how often, which entities were affected, and whether there were errors. This is the “was my automation working while I wasn’t looking?” question.

Bïrch MCP with list_rule_logs returns run-by-run summaries (timestamp, entities evaluated/acted on/errored), and get_rule_log returns per-entity outcomes (which ad was paused, which budget changed, what condition values were evaluated, what errors occurred). Can filter by matched entities, errors only, or specific entity IDs.

An operator or manager on a mature account where rules accumulated over months. Often triggered by unexplained performance instability, a budget spike, or an account handover: “I have 80 rules and I’ve lost track — what’s actually running, and is anything working against me?”

Ask Claude to inventory all rules, check recent execution history, and classify them: firing usefully, enabled but never triggering (stale thresholds or conditions that no longer match), firing too broadly (acting on most entities every run), or fighting each other (one rule pauses what another restarts). If your account is organized into folders, scope the audit to one section at a time.

1. “What happened while I wasn’t looking?” (overnight/weekend briefing)

You need: ad set and ad level performance and trends, change history (what was modified and when), anomaly alerts (spikes, drops, unusual patterns), budget and delivery status across active and paused campaigns. The job: get caught up before the first call.

2. “My automation made changes — were they the right calls?” (rule validation)

You need: entity-level performance data (the ad set that got paused — was it actually underperforming?), auction insights (was it a bid problem or a creative quality problem?), change history (what other changes happened around the same time that might explain the shift). The job: close the loop between what the rules did and what actually happened.

3. “Am I doing the right things compared to others?” (calibration)

You need: industry benchmarks filtered by spend tier/goal, Opportunity Score, Ads Library for competitive creative. The job: zoom out from your own account and check whether your strategy makes sense in context.

You ask “what rules should I set up?” Bïrch AI looks at connected ad accounts, current performance, and existing rules (or lack of them), then suggests rules from the Strategy Library tailored to what it sees. On Meta, it creates clickable draft links. On other platforms, it gives text recommendations.

Also works proactively: when you visit the rules page, Bïrch AI can prompt you with “check new rule ideas for your account” — no need to know to ask.

You describe what you want: “pause my ad sets if spend is over 100 for the last 3 days and cost per purchase is above 40.” Bïrch AI creates a draft rule, saved to the “Birch AI Playbook” folder. You review, edit if needed, and activate.

Draft generation is Meta only — on TikTok, Snapchat and Google you get the rule logic as text to build from.

When you want to ideate rules in Claude: search_metrics searches Bïrch’s metric catalog by natural language (“purchase cost on TikTok,” “frequency for Snapchat”). Returns canonical metric ID, label, platform, and whether it’s usable in automation rules or reports. Supports Meta, Google, TikTok, Snapchat, and attribution platforms (Hyros, AppsFlyer, Wicked Reports, Adjust, Google Analytics).

You can also get a list of available metrics from Bïrch AI.

“How is my ROAS across all ad accounts this week vs last week?” or “Compare spend for Meta and TikTok for Q1.” Bïrch AI answers in natural language, supports custom timeframes (Q1, last 7 days, etc.), and works across all connected platforms at the ad account level. Entity-level breakdown (campaigns, ad sets, ads) is Meta only and limited to accounts with a manageable entity count.

You notice yo-yoing budgets or unstable performance. You ask Bïrch AI to review the rule set for offense/defense imbalance. The AI reads the full setup and identifies patterns: scaling based on single-day data, no budget caps, aggressive percentage increases without guardrails, missing recovery rules for paused winners.

Bïrch AI can correlate with actual account performance. Bïrch MCP can read rule logic via get_rule_details and spot the structural imbalance, but won’t see the performance impact.

You’re not just looking at what happened — you want to know where you’re leaving money on the table. Bïrch AI can identify high-performing ad sets that are underfunded, poor performers that are consuming budget, and recommend scaling or pausing decisions.

For debugging: “why did this rule pause this ad?” Use get_rule_log with specific entity IDs to see the per-entity outcome — which condition was evaluated, what metric value was checked, what action was applied, and any error messages. Also shows the historical rule snapshot at time of execution, so Claude can compare what the rule looked like when it ran vs. what it looks like now.

Search the Facebook Ads Library using plain language: “show me active video ads from [competitor] in the US.” No need to navigate the Ads Library UI. Only covers publicly visible ads — not competitor performance data.

Compare ad account performance against auction and industry benchmarks. This data is only available through Meta MCP — it’s not in the regular Meta API or Bïrch’s own analytics.

Check for unusual spend patterns, sudden metric shifts, and Meta’s Opportunity Score. Also supports listing custom audiences, images, and videos in the account — useful for quick creative audits without navigating Ads Manager.

Bïrch AI can suggest rules for creative fatigue detection (frequency + CTR decay), creative testing pipelines (auto-pause underperformers, scale winners), and late-attribution recovery (“phoenix” rules that restart paused ads when delayed conversions come in). These are text recommendations (all platforms) or draft rules (Meta only).

Check which uploads completed, which failed, and which ads were created. Useful for teams doing high-volume creative testing who want to monitor Stage from Claude without logging into Bïrch. Supports Meta, TikTok, Snapchat.

Connect at mcp.bir.ch/mcp. Claude can then pull workspaces, ad accounts, rules, rule details, execution logs, folder structures, Stage tasks, and help docs — all in natural language. No Bïrch UI needed.

The differentiation from Bïrch AI: no UI constraints, exportable analysis, code execution against the data, and the ability to process very large rule sets (200+ rules). Better for deep audits, cross-account comparisons, and structured exports.

Enterprise teams that run their own AI bots (e.g., in Slack) can connect Bïrch MCP. Read-only by design — the bot reports, humans decide what to change.

search_help searches Bïrch Help Center articles by keyword. Useful when a user asks Claude “how does the ranking condition work in Bïrch?” — Claude can pull the answer from Bïrch’s own documentation.