improvement(search): send repeated account notes and fixed fields once in Search MCP results

Author: waleedlatif1

apps/sim/lib/api/contracts/knowledge/mcp.ts

@@ -19,8 +19,28 @@ export const organizationKnowledgeMcpContract = defineRouteContract({
 
 const LIVE_SEARCH_MCP_DEFAULT_TOP_K = { federated: 30, scoped: 20 } as const
 
+/**
+ * A datetime field without the regex its JSON Schema would otherwise carry beside
+ * `format: "date-time"`. Every tools/list call resends the schema, and parsing still rejects a
+ * string that is not an ISO datetime with an offset.
+ */
+function withoutPatternInJsonSchema<T extends z.ZodType>(schema: T): T {
+  return schema.meta({ pattern: undefined })
+}
+
+const { startDate, endDate, modifiedAfter, modifiedBefore } = workspaceSearchFiltersSchema.shape
+
+/** The date filters of the Search MCP tools, as their tools/list schema presents them. */
+const mcpDateFilterFields = {
+  startDate: withoutPatternInJsonSchema(startDate),
+  endDate: withoutPatternInJsonSchema(endDate),
+  modifiedAfter: withoutPatternInJsonSchema(modifiedAfter),
+  modifiedBefore: withoutPatternInJsonSchema(modifiedBefore),
+}
+
 export const liveSearchMcpSchema = searchWorkspaceInputFieldsSchema
   .extend({
+    ...mcpDateFilterFields,
     topK: searchWorkspaceInputFieldsSchema.shape.topK
       .unwrap()
       .optional()
@@ -81,6 +101,7 @@ export const readLiveDocumentMcpSchema = readDocumentInputSchema
 
 export const chatSearchMcpSchema = workspaceSearchFiltersSchema
   .extend({
+    ...mcpDateFilterFields,
     query: z.string().trim().min(1, 'A question is required').max(8192),
   })
   .strict()

apps/sim/lib/knowledge/mcp/server.protocol.test.ts

@@ -18,6 +18,10 @@ vi.mock('@/lib/api/server/routes/v2-json-route', () => ({
   v2RateLimits: { publicApi: { enforce: vi.fn().mockResolvedValue(null) } },
 }))
 vi.mock('@/lib/sim-search/live/application', () => ({
+  LIVE_MORE_RANKED_BELOW_MESSAGE: 'More matches ranked below.',
+  LIVE_DATE_ORDER_MESSAGE: 'Date order covers retrieved results.',
+  LIVE_NOT_STARTED_MESSAGE: 'Not searched before the deadline.',
+  LIVE_OVER_ACCOUNT_LIMIT_MESSAGE: 'Not searched: over the account limit.',
   searchLiveKnowledge: { execute: hoisted.liveSearch },
   readLiveDocument: { execute: hoisted.liveRead },
 }))
@@ -44,6 +48,7 @@ describe('Search MCP protocol', () => {
     }
     mocks.liveSearch.mockResolvedValue({
       results: [row],
+      retrieval: { status: 'complete', timedOutLegs: [] },
       live: { backend: 'live', accounts: [], guidance: '' },
     })
     mocks.liveRead.mockResolvedValue({
@@ -74,6 +79,8 @@ describe('Search MCP protocol', () => {
       const searchSchema = tools.find((tool) => tool.name === 'search')?.inputSchema
       const readSchema = tools.find((tool) => tool.name === 'read_document')?.inputSchema
       expect(searchSchema?.properties).toHaveProperty('nativeQueries')
+      expect(searchSchema?.properties?.startDate).toMatchObject({ format: 'date-time' })
+      expect(searchSchema?.properties?.startDate).not.toHaveProperty('pattern')
       expect(readSchema?.properties).toHaveProperty('startOffset')
       expect(readSchema?.properties).not.toHaveProperty('url')
       const search = await client.callTool({

apps/sim/lib/knowledge/mcp/server.test.ts

@@ -21,6 +21,12 @@ const hoisted = vi.hoisted(() => ({
   rateLimit: vi.fn(),
   afterResponse: vi.fn<(task: () => Promise<void>) => void>(),
   recordActivity: vi.fn(),
+  statusMessages: {
+    LIVE_MORE_RANKED_BELOW_MESSAGE: 'More matches ranked below.',
+    LIVE_DATE_ORDER_MESSAGE: 'Date order covers retrieved results.',
+    LIVE_NOT_STARTED_MESSAGE: 'Not searched before the deadline.',
+    LIVE_OVER_ACCOUNT_LIMIT_MESSAGE: 'Not searched: over the account limit.',
+  },
 }))
 vi.mock('@/lib/core/utils/after-response', () => ({ afterResponse: hoisted.afterResponse }))
 vi.mock('@/lib/knowledge/mcp/activity', () => ({
@@ -46,6 +52,7 @@ vi.mock('@/lib/api/server/routes/v2-json-route', () => ({
   v2RateLimits: { publicApi: { enforce: hoisted.rateLimit } },
 }))
 vi.mock('@/lib/sim-search/live/application', () => ({
+  ...hoisted.statusMessages,
   searchLiveKnowledge: { execute: hoisted.liveSearch },
   readLiveDocument: { execute: hoisted.liveRead },
 }))
@@ -58,6 +65,7 @@ import { createKnowledgeMcpServer } from '@/lib/knowledge/mcp/server'
 import type { ResolvedSecretTraceRegistry } from '@/executor/utils/resolved-secret-trace-registry'
 
 const mocks = hoisted
+const retrieval = { status: 'complete', timedOutLegs: [] }
 
 urlsMockFns.mockGetBaseUrl.mockReturnValue('https://sim.example')
 
@@ -83,7 +91,7 @@ beforeEach(() => {
   mocks.tools.clear()
   mocks.configs.clear()
   mocks.rateLimit.mockReset().mockResolvedValue(null)
-  mocks.liveSearch.mockResolvedValue({ results: [] })
+  mocks.liveSearch.mockResolvedValue({ results: [], retrieval })
   mocks.chat.mockResolvedValue({ content: 'An answer', citations: [] })
 })
 
@@ -110,7 +118,7 @@ describe('live organization Search MCP', () => {
     mocks.liveSearch.mockResolvedValue({
       results: [document],
       live: coverage,
-      retrieval: { status: 'partial' },
+      retrieval: { status: 'partial', timedOutLegs: [] },
     })
     mocks.liveRead.mockResolvedValue({
       ...document,
@@ -197,7 +205,7 @@ describe('live organization Search MCP', () => {
     mocks.liveSearch.mockResolvedValueOnce({
       results: [message, event],
       live: coverage,
-      retrieval: { status: 'complete' },
+      retrieval,
     })
     const { results } = textOf(await call('search', { query: 'launch' }))
     expect(results).toEqual([
@@ -225,6 +233,123 @@ describe('live organization Search MCP', () => {
     ])
   })
 
+  it('states each repeated account status once and flags the entries that carried it', async () => {
+    create()
+    const {
+      LIVE_MORE_RANKED_BELOW_MESSAGE: moreBelow,
+      LIVE_DATE_ORDER_MESSAGE: dateOrder,
+      LIVE_NOT_STARTED_MESSAGE: deadline,
+      LIVE_OVER_ACCOUNT_LIMIT_MESSAGE: overLimit,
+    } = mocks.statusMessages
+    const emptyPage = 'No readable matches on this page.'
+    const unconnected = 'No connection with this provider is configured.'
+    mocks.liveSearch.mockResolvedValueOnce({
+      results: [],
+      retrieval: { status: 'partial', timedOutLegs: [] },
+      live: {
+        backend: 'live',
+        accounts: [
+          {
+            accountId: 'a1',
+            provider: 'slack',
+            displayName: 'Slack',
+            status: 'partial',
+            message: `${dateOrder} ${moreBelow}`,
+          },
+          {
+            accountId: 'a2',
+            provider: 'gmail',
+            queryIndex: 1,
+            displayName: 'Mail',
+            status: 'partial',
+            message: `Page note. ${dateOrder} ${emptyPage} ${moreBelow}`,
+            nextCursor: 'c2',
+          },
+          {
+            accountId: 'a3',
+            provider: 'notion',
+            displayName: 'Notion',
+            status: 'timeout',
+            message: deadline,
+          },
+          {
+            accountId: 'a4',
+            provider: 'linear',
+            displayName: 'Linear',
+            status: 'unavailable',
+            message: overLimit,
+          },
+          {
+            accountId: '',
+            provider: 'jira',
+            queryIndex: 2,
+            displayName: 'jira',
+            status: 'reconnect',
+            message: unconnected,
+          },
+          { accountId: 'a5', provider: 'jira', displayName: 'jira', status: 'ok' },
+        ],
+        guidance: 'Use nativeQueries to continue.',
+      },
+    })
+    const { live, retrieval: projected } = textOf(await call('search', { query: 'launch' }))
+    expect(projected).toEqual({ status: 'partial' })
+    expect(live).toEqual({
+      notes: {
+        moreBelow,
+        dateOrderPartial: dateOrder,
+        notSearchedByDeadline: deadline,
+        notSearchedOverLimit: overLimit,
+      },
+      accounts: [
+        {
+          accountId: 'a1',
+          displayName: 'Slack',
+          provider: 'slack',
+          status: 'partial',
+          moreBelow: true,
+          dateOrderPartial: true,
+        },
+        {
+          accountId: 'a2',
+          displayName: 'Mail',
+          provider: 'gmail',
+          queryIndex: 1,
+          status: 'partial',
+          nextCursor: 'c2',
+          message: `Page note. ${emptyPage}`,
+          moreBelow: true,
+          dateOrderPartial: true,
+        },
+        {
+          accountId: 'a3',
+          displayName: 'Notion',
+          provider: 'notion',
+          status: 'timeout',
+          notSearchedByDeadline: true,
+        },
+        {
+          accountId: 'a4',
+          displayName: 'Linear',
+          provider: 'linear',
+          status: 'unavailable',
+          notSearchedOverLimit: true,
+        },
+        { provider: 'jira', queryIndex: 2, status: 'reconnect', message: unconnected },
+        { accountId: 'a5', displayName: 'jira', provider: 'jira', status: 'ok' },
+      ],
+      guidance: 'Use nativeQueries to continue.',
+    })
+  })
+
+  it('rejects a date filter that is not an ISO datetime with an offset', () => {
+    create()
+    for (const tool of ['search', 'chat'])
+      expect(() =>
+        mocks.configs.get(tool)?.inputSchema.parse({ query: 'launch', startDate: '2026-09-01' })
+      ).toThrow()
+  })
+
   it('returns a read document title and link once', async () => {
     create()
     const read = textOf(await call('read_document', { documentId }))
@@ -241,7 +366,7 @@ describe('live organization Search MCP', () => {
       backend.mockImplementationOnce(
         async ({ input }: { input: { resultSecretRegistry: ResolvedSecretTraceRegistry } }) => {
           input.resultSecretRegistry.markIncomplete('source-provenance-incomplete')
-          return tool === 'search' ? { results: [document] } : document
+          return tool === 'search' ? { results: [document], retrieval } : document
         }
       )
       const result = await call(tool, tool === 'search' ? { query: 'release' } : { documentId })

apps/sim/lib/knowledge/mcp/server.ts

@@ -11,6 +11,7 @@ import {
   liveSearchMcpSchema,
   readLiveDocumentMcpSchema,
 } from '@/lib/api/contracts/knowledge/mcp'
+import type { WorkspaceKnowledgeSearchData } from '@/lib/api/contracts/mothership-assistant-tools'
 import type { V2ApiKeyAuthContext } from '@/lib/api/server/routes/v2-api-key-auth'
 import { v2RateLimits } from '@/lib/api/server/routes/v2-json-route'
 import { afterResponse } from '@/lib/core/utils/after-response'
@@ -28,7 +29,15 @@ import {
 } from '@/lib/knowledge/mcp/tool-runner'
 import { liveCitationId } from '@/lib/knowledge/search/citation'
 import { toolError } from '@/lib/mcp/tool-result'
-import { readLiveDocument, searchLiveKnowledge } from '@/lib/sim-search/live/application'
+import {
+  LIVE_DATE_ORDER_MESSAGE,
+  LIVE_MORE_RANKED_BELOW_MESSAGE,
+  LIVE_NOT_STARTED_MESSAGE,
+  LIVE_OVER_ACCOUNT_LIMIT_MESSAGE,
+  readLiveDocument,
+  searchLiveKnowledge,
+} from '@/lib/sim-search/live/application'
+import { separateMessages } from '@/lib/sim-search/live/pages'
 import { LiveReadError } from '@/lib/sim-search/live/read-error'
 import { v2CaughtOrchestrationError } from '@/app/api/v2/lib/response'
 import { ResolvedSecretTraceRegistry } from '@/executor/utils/resolved-secret-trace-registry'
@@ -74,6 +83,55 @@ function sourceFacts(row: SourceRow) {
   }
 }
 
+/**
+ * Account status sentences that repeat across `live.accounts`, keyed by the flag that marks an
+ * entry carrying one. The search result states each sentence once in `live.notes` under its flag.
+ */
+const LIVE_ACCOUNT_NOTES = {
+  moreBelow: LIVE_MORE_RANKED_BELOW_MESSAGE,
+  dateOrderPartial: LIVE_DATE_ORDER_MESSAGE,
+  notSearchedByDeadline: LIVE_NOT_STARTED_MESSAGE,
+  notSearchedOverLimit: LIVE_OVER_ACCOUNT_LIMIT_MESSAGE,
+}
+
+/**
+ * Live coverage with each repeated status sentence moved to `live.notes` and flagged on the
+ * entries that carried it. A reconnect entry for a provider with no connection has no account ID
+ * and is named by its provider, so it sends neither. `backend` is always `live` here.
+ */
+function projectLiveCoverage(live: NonNullable<WorkspaceKnowledgeSearchData['live']>) {
+  const notes: Record<string, string> = {}
+  const accounts = live.accounts.map((entry) => {
+    const { accountId, displayName, message, ...status } = entry
+    const { found, rest: remaining } = separateMessages(message, LIVE_ACCOUNT_NOTES)
+    const flags: Record<string, true> = {}
+    for (const flag of found) {
+      flags[flag] = true
+      notes[flag] = LIVE_ACCOUNT_NOTES[flag]
+    }
+    return {
+      ...(accountId ? { accountId } : {}),
+      ...(accountId || displayName !== entry.provider ? { displayName } : {}),
+      ...status,
+      ...(remaining === undefined ? {} : { message: remaining }),
+      ...flags,
+    }
+  })
+  return {
+    ...(Object.keys(notes).length ? { notes } : {}),
+    accounts,
+    guidance: live.guidance,
+  }
+}
+
+/** A search result's retrieval status, without the timed-out legs a live search never has. */
+function projectRetrieval({
+  timedOutLegs,
+  ...retrieval
+}: WorkspaceKnowledgeSearchData['retrieval']) {
+  return timedOutLegs.length ? { ...retrieval, timedOutLegs } : retrieval
+}
+
 /** One read_document result in the shape the tool returns it. */
 function projectReadResult(result: Awaited<ReturnType<typeof readLiveDocument.execute>>) {
   const {
@@ -173,7 +231,7 @@ export function createKnowledgeMcpServer(context: KnowledgeMcpContext): McpServe
     {
       title: 'Search',
       description:
-        'Search this organization’s sources through their live APIs, within your access and the admin’s source settings. Use source and date filters to narrow results, or nativeQueries for provider queries and pagination. Results carry a passage around their match where the source provides text. Inspect live.accounts for provider status and continuation cursors, and live.guidance for query syntax. Results are candidates, not proof of complete coverage. When a passage does not answer the question, use read_document with the exact returned documentId, and cite citationUrl when available.',
+        'Search this organization’s sources through their live APIs, within your access and the admin’s source settings. Use source and date filters to narrow results, or nativeQueries for provider queries and pagination. Results carry a passage around their match where the source provides text. Inspect live.accounts for provider status and continuation cursors (a true flag on an account, such as moreBelow, means the live.notes sentence under that key applies to it), and live.guidance for query syntax. Results are candidates, not proof of complete coverage. When a passage does not answer the question, use read_document with the exact returned documentId, and cite citationUrl when available.',
       inputSchema: liveSearchMcpSchema,
       annotations: KNOWLEDGE_MCP_READ_ONLY,
     },
@@ -206,8 +264,8 @@ export function createKnowledgeMcpServer(context: KnowledgeMcpContext): McpServe
               connectorType: row.connectorType,
               content: row.content,
             })),
-            retrieval: result.retrieval,
-            live: result.live,
+            retrieval: projectRetrieval(result.retrieval),
+            ...(result.live ? { live: projectLiveCoverage(result.live) } : {}),
           },
           registry
         )