@sendinel/mcp-server 1.0.43 → 1.0.44

package/build/tools/contacts.js

@@ -1,7 +1,7 @@
 import { ResourceTemplate } from '@modelcontextprotocol/sdk/server/mcp.js';
 import { CONTACT_ENRICHMENT_PROVIDERS } from '@desler/models';
 import { createContactEnrichmentAdapters, enrichContactFromConnections, } from '@desler/models/contact-enrichment';
-import { buildPreviewAttentionEnvelope, applySegmentRules, ContactPropertyAnomalyLogger, normalizeContactProperties, getAttentionDecision, getContactNeighborhood, getScheduledTouches, } from '@sendinel/domain';
+import { buildPreviewAttentionEnvelope, applySegmentRules, ContactPropertyAnomalyLogger, normalizeContactProperties, getAttentionDecision, getContactNeighborhood, getScheduledTouches, findContactDuplicateGroups, } from '@sendinel/domain';
 import { z } from 'zod';
 import { db, projectClientForWriteback } from '../db.js';
 import { getProjectId } from '../project.js';
@@ -2104,7 +2104,7 @@ Requirements: urgency without desperation, acknowledge the gap, one clear CTA, v
             idempotentHint: true,
             openWorldHint: false,
         },
-        description: 'Find duplicate contacts using fuzzy email matching and exact name matching. Returns grouped candidates with confidence scores.',
+        description: 'Find duplicate contacts using fuzzy email matching and exact name matching. Returns grouped candidates with confidence scores. Each group also suggests which contact to keep (suggested_target_id, source_ids) for merge_subscribers, and flags probably_different_people when the first names differ (e.g. two people at one company): never merge those without asking the user.',
         inputSchema: {
             confidence_threshold: z.number().min(0.5).max(1).optional().describe('Minimum similarity (0.5-1.0, default 0.8)'),
             limit: z.number().int().min(1).max(200).optional().describe('Max duplicate groups (default 50)'),
@@ -2130,15 +2130,35 @@ Requirements: urgency without desperation, acknowledge the gap, one clear CTA, v
                 return j({ duplicate_groups: [], total_groups: 0, coverage: { contacts_scanned: 0, contacts_matching: 0, contact_limit: CONTACT_SCAN_LIMIT, complete: true } });
             siteIds = ids;
         }
+        // If the last sync time can't be read, hide nothing extra: a longer list beats a missing duplicate.
+        let byodCutoff = null;
+        try {
+            const { data: lastSync } = await db()
+                .from('cron_runs')
+                .select('started_at')
+                .eq('job_name', 'byod-sync')
+                .eq('status', 'success')
+                .order('started_at', { ascending: false })
+                .limit(1)
+                .maybeSingle();
+            byodCutoff = lastSync?.started_at ?? null;
+        }
+        catch {
+            byodCutoff = null;
+        }
         const pageQuery = (from) => {
             let q = db()
                 .from('subscribers')
                 .select('id, email, first_name, last_name, engagement_score, lifecycle_stage, created_at, source', { count: 'exact' })
-                .eq('project_id', projectId)
-                .eq('global_unsubscribed', false);
+                .eq('project_id', projectId);
             // #3604: hide seed/fixture rows like every other default contact view.
             if (!include_test)
                 q = q.not('is_test', 'is', true);
+            // #3750: scan the same contacts as the dashboard's Find & Merge Duplicates screen. Unsubscribed
+            // contacts are still duplicates (a merge keeps their consent), and rows the last BYOD sync
+            // no longer saw are hidden there.
+            if (byodCutoff)
+                q = q.or(`source.neq.database_import,updated_at.gte.${byodCutoff}`);
             if (siteIds)
                 q = q.in('id', siteIds);
             return q.order('created_at', { ascending: true }).order('id', { ascending: true })
@@ -2168,82 +2188,20 @@ Requirements: urgency without desperation, acknowledge the gap, one clear CTA, v
         const coverage = { contacts_scanned: scanned, contacts_matching: matching, contact_limit: CONTACT_SCAN_LIMIT, complete: matching === null ? null : matching <= scanned };
         if (!contacts?.length)
             return j({ duplicate_groups: [], total_groups: 0, coverage });
-        const lev = (a, b) => {
-            const m = a.length, n = b.length;
-            const dp = Array.from({ length: m + 1 }, (_, i) => i);
-            for (let jj = 1; jj <= n; jj++) {
-                let prev = dp[0];
-                dp[0] = jj;
-                for (let i = 1; i <= m; i++) {
-                    const tmp = dp[i];
-                    dp[i] = a[i - 1] === b[jj - 1] ? prev : 1 + Math.min(prev, dp[i], dp[i - 1]);
-                    prev = tmp;
-                }
-            }
-            return dp[m];
-        };
-        const similarity = (a, b) => {
-            if (a === b)
-                return 1;
-            const maxLen = Math.max(a.length, b.length);
-            if (maxLen === 0)
-                return 1;
-            return 1 - lev(a, b) / maxLen;
-        };
-        const byDomain = {};
-        for (const c of contacts) {
-            const [, domain] = c.email.toLowerCase().split('@');
-            if (!byDomain[domain])
-                byDomain[domain] = [];
-            byDomain[domain].push(c);
-        }
-        const groups = [];
-        const merged = new Set();
-        for (const domain of Object.keys(byDomain)) {
-            const dc = byDomain[domain];
-            for (let i = 0; i < dc.length; i++) {
-                if (merged.has(dc[i].id))
-                    continue;
-                const localA = dc[i].email.split('@')[0].toLowerCase();
-                const group = [dc[i]];
-                for (let jj = i + 1; jj < dc.length; jj++) {
-                    if (merged.has(dc[jj].id))
-                        continue;
-                    const localB = dc[jj].email.split('@')[0].toLowerCase();
-                    const sim = similarity(localA, localB);
-                    if (sim >= confidence_threshold && sim < 1) {
-                        group.push(dc[jj]);
-                        merged.add(dc[jj].id);
-                    }
-                }
-                if (group.length > 1) {
-                    merged.add(dc[i].id);
-                    const avgSim = similarity(group[0].email.split('@')[0].toLowerCase(), group[1].email.split('@')[0].toLowerCase());
-                    groups.push({ group, confidence: Math.round(avgSim * 100) / 100, reason: 'similar_email' });
-                }
-            }
-        }
-        const byName = {};
-        for (const c of contacts) {
-            if (merged.has(c.id))
-                continue;
-            const fn = (c.first_name ?? '').toLowerCase().trim();
-            const ln = (c.last_name ?? '').toLowerCase().trim();
-            if (!fn && !ln)
-                continue;
-            const key = `${fn}|${ln}`;
-            if (!byName[key])
-                byName[key] = [];
-            byName[key].push(c);
-        }
-        for (const nameGroup of Object.values(byName)) {
-            if (nameGroup.length > 1) {
-                groups.push({ group: nameGroup, confidence: 0.7, reason: 'same_name' });
-            }
-        }
-        groups.sort((a, b) => b.confidence - a.confidence);
+        // The matching rule is shared with the dashboard screen (@sendinel/domain, #3743/#3750).
+        const groups = findContactDuplicateGroups(contacts, { threshold: confidence_threshold })
+            .map((g) => ({
+            group: g.contacts,
+            confidence: g.confidence,
+            reason: g.reason,
+            suggested_target_id: g.suggested_target_id,
+            source_ids: g.source_ids,
+            probably_different_people: g.probably_different_people,
+            different_first_names: g.different_first_names,
+        }));
         const result = groups.slice(0, maxGroups);
-        return j({ duplicate_groups: result, total_groups: result.length, coverage });
+        // total_groups counts every group found; groups_truncated says when limit cut the list.
+        return j({ duplicate_groups: result, total_groups: groups.length, groups_truncated: groups.length > result.length, coverage });
     });
     server.registerTool('merge_subscribers', {
         title: 'Merge Subscribers',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sendinel/mcp-server",
-  "version": "1.0.43",
+  "version": "1.0.44",
   "description": "AI-controlled email operations control plane for campaigns, contacts, segments, analytics, and multi-provider sending via MCP.",
   "mcpName": "ai.sendinel/mcp-server",
   "type": "module",