Members

# constant AGENT_TRIGGERS

Agent trigger patterns - complexity indicators that require reasoning If any of these match, route to the agent

# constant ActionTypes

MCP Action Types

Contract between MCP Server and Frontend. When MCP executes a tool, it returns an action object that tells the frontend what client-side action to take.

# constant CANCEL_PHRASES

Phrases that indicate user wants to cancel/abandon a pending action

# constant CLASSIFY_PATTERNS

Patterns that suggest simple pass-through to intent classifier These are ambiguous but likely simple

# constant ConfirmationTypes

Confirmation types for CONFIRMATION_REQUIRED action

# constant DATABASE_SCHEMAS

Database schema definitions for AI query generation These are simplified schemas that tell AI what fields are queryable

# constant DEFAULT_CONFIG

Default configuration

# constant DIRECT_ACTION_PATTERNS

Direct action patterns - high confidence, single step These bypass any AI reasoning and execute immediately

# constant EXTRACTION_PROMPTS

Extraction type prompts

# constant FILE_OPEN_TRIGGERS

Direct file open patterns - single file open action These need agent routing because they require finding the file first, then returning VIEW_FILE action

# constant FILE_QUERY_TRIGGERS

File query patterns - listing/finding files (whiteboards, notes, documents) These need agent routing because they require query_files with specific parameters

# constant FOLLOW_UP_TEMPLATES

Follow-up question templates for missing parameters

# constant PRIMITIVE_TOOLS

Tool metadata for agent orchestrator Maps tool names to their implementations and descriptions

# constant REQUIRED_PARAMS

Required parameters for action tools These must be collected before execution

# constant RouteType

Route types

# constant SIMPLE_QUERY_PATTERNS

Simple query patterns - single database fetch These use the existing query_data tool

# constant SUMMARY_PROMPTS

Summary type prompts

# constant StepType

Agent step types (for reasoning display)

# constant TRANSCRIPT_VIEW_TRIGGERS

Direct transcript view patterns - VIEW/SHOW a specific transcript These route to agent which uses get_transcript tool (returns VIEW_TRANSCRIPT action)

# _aiServiceInstance

Cached UnifiedAIService instance (singleton)

# constant buildSecurityContext

Build security context for MCP tool calls

Methods

# _buildMeetingsGraph()

Build a star graph of you → distinct meeting rooms.

Room records in the history are per-session — the same room name can appear hundreds of times (each session = one Room document). Showing every session as its own node is noisy AND causes node-id collisions in the force graph. We dedupe by name here: one node per distinct room name, with the session count and the most-recent session date carried on properties. Drilling into a room node fetches that room's participants.

# _buildRecordingsGraph()

Build a star graph of you → distinct recorded rooms.

Same dedup strategy as _buildMeetingsGraph: multiple recordings often share the same roomName (one recording per session of a re-used room). Showing every recording as a separate node is noisy and causes node-id collisions in the force graph, which orphans most nodes from "You". Dedupe by roomName, carry a recording count + latest date on properties.

# async _expandDateGroup()

Expand a date-group node into the individual sessions of a room on that day.

# async _expandMeeting()

Expand a meeting node into its participants.

# async _expandPerson()

Expand a person node.

If the person IS the current user → list all meetings they attended.
Otherwise → list meetings the two of them SHARED (we can't query another user's full meeting history without org-owner auth; shared meetings are data the current user already participated in, so they're safe to read).

# async _expandRoomSessions()

Expand a deduped room node (sessionCount > 1) into per-day session nodes.

One node per distinct day on which a session of this room occurred.
If a day has exactly one session: that node is a regular meeting node with the session's roomId — drilling shows participants.
If a day has multiple sessions: that node is a dategroup — drilling shows the individual sessions.

# async _expandSummary()

Expand a stats summary node into the actual entities of that type.

# addBreadcrumbLabel()

Helper to append a breadcrumbLabel to a createGraphResponse result.

# analyzeQuery()

Get routing analysis for a query Useful for debugging

# answerQuestion(params) → {Promise.<Object>}

Answer a specific question about provided text

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`text`	`string`	Text to answer question about
`question`	`string`	Question to answer
`citeSources`	`boolean`	Include citations from text (default: true)

Returns:

Answer result

Type: Promise.<Object>

# applyAuthorizationFilters(query, userContext, type) → {Object}

Apply authorization filters to a query based on user context

Parameters:

Name	Type	Description
`query`	`Object`	MongoDB query object
`userContext`	`Object`	User context with apiKey
`type`	`string`	Resource type ('room', 'schedule')

Returns:

Modified query with authorization filters

Type: Object

# buildUserContext(user) → {Object}

Build authorization context for tools Extracts and normalizes user information from request

Parameters:

Name	Type	Description
`user`	`Object`	User object from authentication middleware

Returns:

Normalized user context

Type: Object

# calculateNextOccurrence()

Calculate next occurrence for recurring schedules

# async callAI()

Call AI API

# cancelSchedule(params, securityContext) → {Promise.<Object>}

Cancel a scheduled meeting

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`scheduleId`	`string`	Schedule ID to cancel (required)
`reason`	`string`	Cancellation reason
`notifyParticipants`	`boolean`	Send cancellation emails (default: true)
`securityContext`	`Object`	User security context

Returns:

Result with action for frontend

Type: Promise.<Object>

# checkTranscriptions(params, securityContext) → {Promise.<Object>}

Check which rooms have transcription data

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomNames`	`Array.<string>`	Specific rooms to check (optional)
`limit`	`number`	Max rooms to return (default: 20)
`recentFirst`	`boolean`	Sort by most recent transcription
`securityContext`	`Object`	User security context

Returns:

Rooms with transcription status

Type: Promise.<Object>

# cleanParticipantName()

Clean participant name by removing timestamp (prefix or suffix) If name is "Unknown", fall back to participant_id with timestamp removed e.g., "1765915346271_its_me_again" → "its_me_again" e.g., "karl_1734567890123" → "karl"

# createDbAdapter()

Create a minimal dbConnection adapter for GraphContextService Uses models from dependency injection instead of raw mongoose.connection.db

# createMCPServer(configopt) → {Object}

Create and initialize the MCP Server

Models and AI services are bundled internally. Callers only need to provide aiConfig (API keys) and ensure the database connection is established.

Parameters:

Name	Type	Attributes	Description
`config`	`Object`	<optional>	Configuration object
`models`	`Object`	<optional>	Override default models from muziertc-models
`utils`	`Object`	<optional>	Optional utility overrides
`aiConfig`	`Object`		AI provider configuration
`aiConfig.openaiKey`	`string`	<optional>	OpenAI API key
`aiConfig.xaiKey`	`string`	<optional>	xAI/Grok API key

Returns:

MCP server components

Type: Object

# createRoom(params, securityContext) → {Promise.<Object>}

Create a new persistent room

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomName`	`string`	Name for the room (required)
`password`	`string`	Password for the room (optional - if provided, usePassword is auto-enabled)
`usePassword`	`boolean`	Require password to join
`useWaitingRoom`	`boolean`	Enable waiting room
`enableOffline`	`boolean`	Enable offline features (default: true)
`autoIntelligence`	`boolean`	Auto-start transcription/recording (default: false)
`securityContext`	`Object`	User security context

Returns:

Result with action for frontend

Type: Promise.<Object>

# createSchedule(params, securityContext) → {Promise.<Object>}

Create a scheduled meeting

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomName`	`string`	Meeting name (required)
`startDateTime`	`string`	Start date/time ISO string (required)
`endDateTime`	`string`	End date/time (optional)
`duration`	`number`	Duration in minutes (default: 60)
`participants`	`Array.<string>`	Email addresses to invite
`description`	`string`	Meeting description
`recurrence`	`Object`	Recurrence pattern { type, interval, count }
`securityContext`	`Object`	User security context

Returns:

Result with action for frontend

Type: Promise.<Object>

# ensureAIClient()

Initialize AI client (lazy) Uses shared getAIClient utility for OpenAI/xAI support

# async executeAgent()

Execute via agent orchestrator (complex queries)

# async executeDirect()

Execute a direct action (fast path)

# async executeSimple()

Execute a simple query (single database fetch)

# expandGraphNode(params, securityContext) → {Promise.<Object>}

Expand a graph node into its next level.

Hierarchy: stats summary node (summary_person, summary_meeting, summary_recording, summary_topic) → the actual entities of that type person node → meetings that person attended meeting node → participants of that meeting

Parameters:

Name	Type	Description
`params`	`Object`
`nodeId`	`string`	The graph node id (e.g. "summary_meeting", "person_alice_at_x_com")
`nodeType`	`string`	Node type: 'person' \| 'meeting' \| 'recording' \| 'topic' \| 'summary'
`properties`	`Object`	Node properties (email, roomName, entityType, ...)
`limit`	`number`	Max children to return (default: 20)
`securityContext`	`Object`

Returns:

{ success, graph: { nodes, edges, metadata }, graphType, breadcrumbLabel }

Type: Promise.<Object>

# extractDisplayNameFromEmail()

Extract display name from email (must match graph-visualization.js)

# extractInfo(params) → {Promise.<Object>}

Extract structured information from text

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`text`	`string`	Text to extract from
`extractType`	`string`	Type of extraction: 'action-items', 'decisions', 'participants', 'dates', 'topics', 'questions', 'sentiment'
`context`	`string`	Optional context about the text

Returns:

Extraction result

Type: Promise.<Object>

# filterOutBots()

Filter out bot users from participant arrays Bot prefixes: recorder_, rectrans_, streamer_, transcriber_

# findCollaborators(params, securityContext) → {Promise.<Object>}

Find collaborators for a person based on shared meeting attendance

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`personEmail`	`string`	Email of person to find collaborators for (optional, defaults to current user)
`minMeetings`	`number`	Minimum shared meetings (default: 1)
`timeframe`	`string`	Time range: 'last_week', 'last_month', 'last_quarter', 'last_year'
`startDate`	`string` \| `Date`	Specific start date for filtering (ISO string or Date)
`endDate`	`string` \| `Date`	Specific end date for filtering (ISO string or Date)
`limit`	`number`	Max results (default: 20)
`includeTeam`	`boolean`	Include team members' data (org owners only)
`excludeSelf`	`boolean`	Exclude the requesting user from results (default: true)
`securityContext`	`Object`	User security context

Returns:

Collaborator list

Type: Promise.<Object>

# findRecordingsByTopic(params, securityContext) → {Promise.<Object>}

Find recordings that cover a specific topic (via COVERS relationships)

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`topic`	`string`	Topic name to search for
`minConfidence`	`number`	Minimum confidence (0-1, default: 0.5)
`limit`	`number`	Max results (default: 20)
`includeTeam`	`boolean`	Include team members' data (org owners only)
`securityContext`	`Object`	User security context

Returns:

Matching recordings

Type: Promise.<Object>

# findRelatedTopics(params, securityContext) → {Promise.<Object>}

Find topics related to a given topic via shared meetings

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`topic`	`string`	Topic name or partial match
`maxHops`	`number`	Maximum graph traversal depth (default: 2)
`limit`	`number`	Max results (default: 20)
`includeTeam`	`boolean`	Include team members' data (org owners only)
`securityContext`	`Object`	User security context

Returns:

# findSimilarTopics(params, securityContext) → {Promise.<Object>}

Find semantically similar topics using embedding vectors

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`query`	`string`	Text to find similar topics for (required)
`limit`	`number`	Max results (default: 10)
`minSimilarity`	`number`	Minimum similarity threshold 0-1 (default: 0.7)
`securityContext`	`Object`	User security context

Returns:

# async findStoredRoom(roomName, cleanedOwner, apiKey) → {Promise.<(Object|null)>}

Find a stored room by name (exact match or base name without suffix)

Parameters:

Name	Type	Description
`roomName`	`string`	Room name to search for
`cleanedOwner`	`string`	Cleaned owner user ID
`apiKey`	`string`	Organization API key

Returns:

Stored room or null

Type: Promise.<(Object|null)>

# formatDate()

Format date for display

# formatDuration()

Format duration in minutes to human-readable string

# formatExamplesForPrompt()

Format examples from JSON into prompt text

# formatMeetingDate()

Format a date as a short label (e.g. "May 14, 2026")

# formatStoredSummary()

Format a stored transcription summary (ai_analysis or openai_analysis) as readable text Returns null if the summary is incomplete or missing

# formatTranscriptWithSentiment()

Format transcriptions with sentiment data for AI context

# generateResponse(params) → {Promise.<Object>}

Generate a user-friendly response from structured data

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`query`	`string`	Original user query
`data`	`Object` \| `Array`	Data to format into response
`dataType`	`string`	Type of data: 'rooms', 'schedules', 'files', 'search_results', etc.
`format`	`string`	Output format: 'text', 'markdown', 'table' (default: 'markdown')

Returns:

Formatted response

Type: Promise.<Object>

# getAIService()

Get or create a configured UnifiedAIService instance (cached singleton) Uses dependency injection for API keys

# getAccessibleRoomIds(userId, apiKey) → {Promise.<Array.<string>>}

Get IDs of rooms the user has access to (owns or participates in)

Parameters:

Name	Type	Description
`userId`	`string`	User's email or clean user ID
`apiKey`	`string`	User's apiKey for organization filtering

Returns:

Array of accessible room IDs

Type: Promise.<Array.<string>>

# getAccessibleScheduleIds(userId, apiKey) → {Promise.<Array.<string>>}

Get IDs of schedules the user has access to (organizer or invited)

Parameters:

Name	Type	Description
`userId`	`string`	User's email
`apiKey`	`string`	User's apiKey for organization filtering

Returns:

Array of accessible schedule IDs

Type: Promise.<Array.<string>>

# getCollaboratorGraph(params, securityContext) → {Promise.<Object>}

Get collaborator network as a visualization-ready graph Returns both summary text and graph data for force-directed visualization

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters (same as findCollaborators)
`securityContext`	`Object`	User security context

Returns:

Response with summary, graph, and VIEW_GRAPH action

Type: Promise.<Object>

# getDependencies() → {Object}

Get current dependencies Used by MCP components to access models, services, utils

Throws:

If dependencies not initialized

Type: Error

Returns:

Dependencies object

Type: Object

# getEmbeddingService()

Helper to get embedding service from dependency injection

# getFollowUpQuestion()

Get follow-up question for tools that need more info

# getGraphContextService()

Get GraphContextService from dependencies

# getGraphStats(params, securityContext) → {Promise.<Object>}

Get statistics about the knowledge graph

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`includeTeam`	`boolean`	Include team members' data (org owners only)
`securityContext`	`Object`	User security context

Returns:

Graph statistics

Type: Promise.<Object>

# getMeetingTopics(params, securityContext) → {Promise.<Object>}

Get topics discussed in a specific meeting

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`roomName`	`string`	Room/meeting name
`roomId`	`string`	Room ID (alternative to roomName)
`includeTeam`	`boolean`	Include team members' data (org owners only)
`securityContext`	`Object`	User security context

Returns:

Meeting topics

Type: Promise.<Object>

# getModels()

Helper to get models from dependency injection

# getNetworkStatsGraph(params, securityContext) → {Promise.<Object>}

Get network statistics as a visualization-ready graph Shows entity counts around the current user

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`securityContext`	`Object`	User security context

Returns:

Response with summary, graph, and VIEW_GRAPH action

Type: Promise.<Object>

# getOpenAIToolDefinitions()

Format tools for OpenAI function calling

# getPersonMeetings(params, securityContext) → {Promise.<Object>}

Get meetings attended by a person

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`personEmail`	`string`	Email of person (optional, defaults to current user)
`limit`	`number`	Max results (default: 50)
`includeTeam`	`boolean`	Include team members' data (org owners only)
`securityContext`	`Object`	User security context

Returns:

Meeting list

Type: Promise.<Object>

# getRecordingParticipants(params, securityContext) → {Promise.<Object>}

Get all participants (speakers) in a specific recording

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`recordingId`	`string`	Recording ID
`includeTeam`	`boolean`	Include team members' data (org owners only)
`securityContext`	`Object`	User security context

Returns:

Participant list with speaking stats

Type: Promise.<Object>

# getRecordingsForPerson(params, securityContext) → {Promise.<Object>}

Get recordings where a person spoke (via SPOKE_IN relationships)

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`personEmail`	`string`	Email of person (optional, defaults to current user)
`limit`	`number`	Max results (default: 20)
`includeTeam`	`boolean`	Include team members' data (org owners only)
`securityContext`	`Object`	User security context

Returns:

Recording list

Type: Promise.<Object>

# getStructuredSummaryPrompt()

Get a structured JSON prompt for generating cacheable summaries Returns JSON that matches the openai_analysis format used by existing viewers

# getSummaryPrompt()

Get summary prompt based on summary type CRITICAL: All prompts explicitly prevent hallucination

# getToolCategories()

Get tool categories

# getToolDefinition()

Get tool definition by name

# getTopicExtractionService()

Get TopicExtractionService from dependencies

# getTranscript(params, securityContext) → {Promise.<Object>}

Get transcript content for a room

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomName`	`string`	Room name to get transcript for
`roomId`	`string`	Alternatively, room ID
`meetingOffset`	`number`	Skip N most recent meetings (0=most recent, 1=second to last, etc.)
`limit`	`number`	Max transcript entries (default: 500)
`format`	`string`	'raw' (array) or 'formatted' (text with speakers)
`securityContext`	`Object`	User security context

Returns:

Transcript data

Type: Promise.<Object>

# getTranscriptionModel()

Helper to get Transcription model - prefer this over raw collection access

# getUtils()

Helper to get utils from dependency injection

# handleAgentQuery(query, userContext, options) → {Promise.<Object>}

Handle a user query using the agent system

This is the main entry point for the new agent architecture. It uses SmartRouter to determine the best execution path.

Parameters:

Name	Type	Description
`query`	`string`	User's natural language query
`userContext`	`Object`	User context from MCP client
`options`	`Object`	Execution options
`showReasoning`	`boolean`	Include reasoning steps
`forceAgent`	`boolean`	Force agent execution (skip routing)
`fallbackHandler`	`function`	Fallback for classify route

Returns:

Response with summary, action, and optional steps

Type: Promise.<Object>

# hasInitialized() → {boolean}

Check if dependencies have been initialized

Returns:

Type: boolean

# highlightMatch()

Highlight matching text in search results

# isValidEmail(email) → {boolean}

Validate email address format

Parameters:

Name	Type	Description
`email`	`string`	Email to validate

Returns:

True if valid email format

Type: boolean

# isValidTimezone(timezone) → {boolean}

Validate IANA timezone string

Parameters:

Name	Type	Description
`timezone`	`string`	Timezone to validate

Returns:

True if valid timezone

Type: boolean

# loadQueryExamples()

Load query examples from JSON file (lazy loaded)

# normalizeResultRoomNames()

Normalize room names in query results for cleaner display

# normalizeRoomName()

Normalize room name by stripping the alias suffix (random chars after last dash) Only strips if the suffix looks like a generated alias (7 lowercase alphanumeric chars) e.g., "my-room-abc1234" -> "my-room" (stripped - matches alias pattern) e.g., "team-standup" -> "team-standup" (kept - "standup" is not 7 chars)

# openFile(params, securityContext) → {Promise.<Object>}

Open a file for viewing - returns file metadata and VIEW_FILE action

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`resourceType`	`string`	Type of file: 'usernote', 'assignment', 'image', 'recording', etc.
`roomName`	`string`	(Optional) Filter by room name
`fileName`	`string`	(Optional) Filter by file name pattern
`offset`	`number`	Skip N files (0=most recent, 1=second to last, etc.)
`order`	`string`	'newest' (default) or 'oldest'
`securityContext`	`Object`	User security context

Returns:

File data with VIEW_FILE action

Type: Promise.<Object>

# parseRelativeDate()

Parse relative date strings to Date objects

# prettifyRoomName()

Prettify room name for display

Removes random suffix (6-10 alphanumeric chars after last hyphen)
Converts hyphens/underscores to spaces
Applies Title Case

# prettifyRoomNamesInResponse()

Post-process AI response to prettify room names Looks for patterns like "room-name-abc123" and prettifies them Handles room names in markdown bold (name), quotes, or plain text

# queryFiles(params, securityContext) → {Promise.<Object>}

Query files and documents

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`resourceType`	`string`	Filter by type (whiteboard, document, live-transcript, etc.)
`roomName`	`string`	Filter by room
`fileName`	`string`	Filter by file name (partial match)
`foldersOnly`	`boolean`	Only return folders
`filesOnly`	`boolean`	Exclude folders
`sort`	`Object`	Sort specification
`limit`	`number`	Max results
`securityContext`	`Object`	User security context

Returns:

Query results

Type: Promise.<Object>

# queryMeetingContext(params, securityContext) → {Promise.<Object>}

Query meeting mood/sentiment using live context approach

This function:

Loads transcriptions for the specified room/recording (includes sentiment data)
Creates an AI session with transcript+sentiment context
Queries the AI with the user's question

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomName`	`string`	Room name to query
`recordingId`	`string`	Optional specific recording ID
`query`	`string`	User's question about mood/sentiment
`securityContext`	`Object`	User security context

Returns:

AI response with mood analysis

Type: Promise.<Object>

# queryRecordings(params, securityContext) → {Promise.<Object>}

Query recordings

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`roomName`	`string`	Filter by room name
`status`	`string`	Filter by status
`compositionStatus`	`string`	Filter by composition status
`sort`	`Object`	Sort specification
`limit`	`number`	Max results
`securityContext`	`Object`	User security context

Returns:

Query results

Type: Promise.<Object>

# queryRooms(params, securityContext) → {Promise.<Object>}

Query room/meeting history with filters

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`filter`	`Object`	MongoDB filter object (optional)
`roomName`	`string`	Filter by room name (partial match)
`participants`	`Array.<string>`	Filter by participant emails
`startAfter`	`string`	Filter meetings after this date
`startBefore`	`string`	Filter meetings before this date
`completedOnly`	`boolean`	Only return completed meetings (timeEnded != null)
`activeOnly`	`boolean`	Only return active meetings
`sort`	`Object`	Sort specification (default: { timeStarted: -1 })
`limit`	`number`	Max results (default: 20, max: 100)
`computeDuration`	`boolean`	Include computed duration field
`securityContext`	`Object`	User security context

Returns:

Query results

Type: Promise.<Object>

# async queryRoomsForExpansion()

Fetch recent rooms for the current user (used by summary_meeting expansion). Mirrors queryRooms behavior but scoped to expansion needs to keep this primitive self-contained (no circular dep on data-queries).

Room.userIds stores CLEANED user IDs (e.g. "karlreti_t3@gmail_com"), so we must clean the email before querying — otherwise the filter never matches.

# querySchedules(params, securityContext) → {Promise.<Object>}

Query scheduled meetings

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`filter`	`Object`	MongoDB filter object (optional)
`roomName`	`string`	Filter by room name (partial match)
`upcoming`	`boolean`	Only return upcoming schedules
`includePast`	`boolean`	Include past schedules (default: false for upcoming)
`includeRecurring`	`boolean`	Include recurring schedules even if start date is past
`status`	`string`	Filter by status (pending, sent, cancelled)
`sort`	`Object`	Sort specification
`limit`	`number`	Max results (default: 20)
`securityContext`	`Object`	User security context

Returns:

Query results

Type: Promise.<Object>

# queryStoredRooms(params, securityContext) → {Promise.<Object>}

Query user's saved/stored rooms

Parameters:

Name	Type	Description
`params`	`Object`	Query parameters
`roomName`	`string`	Filter by room name (partial match)
`sort`	`Object`	Sort specification
`limit`	`number`	Max results
`securityContext`	`Object`	User security context

Returns:

Query results

Type: Promise.<Object>

# async rehydrateAndQuery()

Create AI session with mood context and query

# async rehydrateAndSummarize()

Rehydrate a transcript by creating a new AI session and pushing the data Also caches the structured summary to the Recording document for future use

# requestInput(params, securityContext) → {Promise.<Object>}

Request input from user via a dynamic form AI uses this when it needs more information to complete an action

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`message`	`string`	Message explaining what's needed
`targetTool`	`string`	Tool to call after collecting input (create_schedule, start_meeting, etc.)
`fields`	`Array`	Fields to collect: [{ name, label, type, required, placeholder }] type: 'text', 'datetime', 'duration', 'emails', 'select'
`prefilled`	`Object`	Values already extracted (will be passed to target tool)
`securityContext`	`Object`	User security context

Returns:

NEEDS_INFO action for frontend

Type: Promise.<Object>

# resolvePersonQuery(requestedEmail, currentUserEmail, isOrgOwner) → {string}

Resolve the personEmail the AI agent supplied into the effective account to query, then authorize it. The agent frequently passes a self-reference in a form other than the raw account email — the cleaned userId form fed to it via room/participant context, or the user's first name / nickname. A non-org-owner can only ever query themselves, so any such reference (and any bare name, which is never a cross-account email) resolves to the current user instead of being mistaken for someone else and denied.

Parameters:

Name	Type	Description
`requestedEmail`	`string` \| `undefined`	personEmail from the tool params
`currentUserEmail`	`string`	the requesting user's account email
`isOrgOwner`	`boolean`	whether the requester is an org owner

Throws:

if a non-owner targets a concrete other account

Type: MCPAuthorizationError

Returns:

the effective email to query

Type: string

# sanitizeNodeId()

Sanitize a string for use as a node ID (must match graph-visualization.js)

# searchContent(params, securityContext) → {Promise.<Object>}

Search through meeting transcriptions, chats, documents, and topics using semantic similarity

Parameters:

Name	Type	Description
`params`	`Object`	Search parameters
`query`	`string`	Search query (what to find)
`contentTypes`	`Array.<string>`	Types to search: 'transcriptions', 'chats', 'documents', 'topics'
`roomName`	`string`	Filter to specific room name
`roomId`	`string`	Filter to specific room ID
`roomNames`	`Array.<string>`	Filter to multiple rooms
`startAfter`	`string`	Filter content after this date
`startBefore`	`string`	Filter content before this date
`limit`	`number`	Max results per type (default: 10)
`minScore`	`number`	Minimum relevance score (default: 0.6)
`securityContext`	`Object`	User security context

Returns:

Search results

Type: Promise.<Object>

# searchTranscriptions(params, securityContext) → {Promise.<Object>}

Simple text search in transcriptions (faster than semantic search)

Parameters:

Name	Type	Description
`params`	`Object`	Search parameters
`query`	`string`	Text to search for
`roomName`	`string`	Filter to specific room
`speaker`	`string`	Filter to specific speaker
`limit`	`number`	Max results (default: 50)
`securityContext`	`Object`	User security context

Returns:

Search results

Type: Promise.<Object>

# sendInvites(params, securityContext) → {Promise.<Object>}

Send meeting invites to participants (This is typically handled by frontend, but can be triggered as action)

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomLink`	`string`	Room link to share
`participants`	`Array.<string>`	Email addresses to invite
`message`	`string`	Optional custom message
`securityContext`	`Object`	User security context

Returns:

Result with action for frontend

Type: Promise.<Object>

# setDependencies(deps)

Set dependencies for use by all MCP components Called by createMCPServer during initialization

Parameters:

Name	Type	Description
`deps`	`Object`	Dependencies object

# shouldUseAgent()

Check if a query should use the agent Can be used by external code to pre-check

# startMeeting(params, securityContext) → {Promise.<Object>}

Start a meeting room

Two modes:

Instant meeting (no roomName) - frontend creates room via InstantRoomDialog flow
Existing room (with roomName) - frontend starts existing room from RoomsList

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomName`	`string`	Room name (optional - if not provided, starts instant meeting)
`participants`	`Array.<string>`	Email addresses to invite
`password`	`string`	Password for the room (optional)
`copyLinkToClipboard`	`boolean`	Copy room link to clipboard
`securityContext`	`Object`	User security context

Returns:

Result with action for frontend

Type: Promise.<Object>

# summarizeText(params) → {Promise.<Object>}

Generate AI summary of provided text

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`text`	`string`	Text to summarize (transcript, notes, etc.)
`summaryType`	`string`	Type of summary: 'meeting-summary', 'action-items', 'key-decisions', 'full', 'brief'
`context`	`string`	Optional context about the text
`maxLength`	`number`	Max output length in tokens (default: 1000)

Returns:

Summary result

Type: Promise.<Object>

# summarizeTranscript(params, securityContext) → {Promise.<Object>}

Summarize a meeting transcript using stored AI context (responseId)

This tool avoids passing large transcript text through JSON by:

Using stored responseId from OpenAI Responses API (persists ~30 days)
Falling back to rehydration if responseId is expired

Parameters:

Name	Type	Description
`params`	`Object`	Parameters
`roomName`	`string`	Room name to get transcript for
`summaryType`	`string`	Type: 'meeting-summary', 'action-items', 'key-decisions', 'full'
`securityContext`	`Object`	User security context

Returns:

Summary result

Type: Promise.<Object>

# validateISODate(dateStr) → {Object}

Validate ISO 8601 date string

Parameters:

Name	Type	Description
`dateStr`	`string`	Date string to validate

Returns:

{ valid: boolean, date: Date|null, error: string|null }

Type: Object

# validatePersonQueryAccess(requestedEmail, currentUserEmail, isOrgOwner)

Validate that the user is authorized to query another person's data. Only org owners can query other users' data within their organization.

Both emails are normalized with cleanUserId (not just lowercased) before comparison. This is the identity form used everywhere else in the system (Room.owner, participant lists, the identity fed to the AI agent), so a self-query expressed in the cleaned form — e.g. the agent passing karlreti_s4@gmail_com when the account email is karlreti+s4@gmail.com — is correctly recognized as the current user instead of being denied. Genuinely different accounts still clean to different ids, so cross-user access remains blocked for non-owners.

Parameters:

Name	Type	Description
`requestedEmail`	`string`	The email being queried
`currentUserEmail`	`string`	The requesting user's email
`isOrgOwner`	`boolean`	Whether the requesting user is an org owner

Throws:

If user is not authorized to query the requested email

Type: MCPAuthorizationError

# validateRoomAccess(roomId, userContext) → {Promise.<Object>}

Check if user has access to a specific room

Parameters:

Name	Type	Description
`roomId`	`string`	Room ID to check
`userContext`	`Object`	User context with userId, email, and apiKey

Throws:

If user doesn't have access

Type: MCPAuthorizationError

Returns:

Room document if user has access

Type: Promise.<Object>

# validateRoomOwnership(roomId, userContext) → {Promise.<Object>}

Check if user owns a room (required for sensitive operations like transcriptions)

Parameters:

Name	Type	Description
`roomId`	`string`	Room ID to check
`userContext`	`Object`	User context with userId, email, and apiKey

Throws:

If user is not the owner

Type: MCPAuthorizationError

Returns:

Room document if user is owner

Type: Promise.<Object>

# validateScheduleAccess(scheduleId, userContext) → {Promise.<Object>}

Validate user access to a schedule

Parameters:

Name	Type	Description
`scheduleId`	`string`	Schedule ID to check
`userContext`	`Object`	User context with userId, email, and apiKey

Throws:

If user doesn't have access

Type: MCPAuthorizationError

Returns:

Schedule document if user has access

Type: Promise.<Object>

# validateScheduleModifyAccess(scheduleId, userContext) → {Promise.<Object>}

Validate user can modify a schedule (must be organizer or admin)

Parameters:

Name	Type	Description
`scheduleId`	`string`	Schedule ID to check
`userContext`	`Object`	User context with userId, email, and apiKey

Throws:

If user can't modify

Type: MCPAuthorizationError

Returns:

Schedule document if user can modify

Type: Promise.<Object>

Hiyve MCP Server

Members

# constant AGENT_TRIGGERS

# constant ActionTypes

# constant CANCEL_PHRASES

# constant CLASSIFY_PATTERNS

# constant ConfirmationTypes

# constant DATABASE_SCHEMAS

# constant DEFAULT_CONFIG

# constant DIRECT_ACTION_PATTERNS

# constant EXTRACTION_PROMPTS

# constant FILE_OPEN_TRIGGERS

# constant FILE_QUERY_TRIGGERS

# constant FOLLOW_UP_TEMPLATES

# constant PRIMITIVE_TOOLS

# constant REQUIRED_PARAMS

# constant RouteType

# constant SIMPLE_QUERY_PATTERNS

# constant SUMMARY_PROMPTS

# constant StepType

# constant TRANSCRIPT_VIEW_TRIGGERS

# _aiServiceInstance

# constant buildSecurityContext

Methods

# _buildMeetingsGraph()

# _buildRecordingsGraph()

# async _expandDateGroup()

# async _expandMeeting()

# async _expandPerson()

# async _expandRoomSessions()

# async _expandSummary()

# addBreadcrumbLabel()

# analyzeQuery()

# answerQuestion(params) → {Promise.<Object>}

Parameters:

# applyAuthorizationFilters(query, userContext, type) → {Object}

Parameters:

# buildUserContext(user) → {Object}

Parameters:

# calculateNextOccurrence()

# async callAI()

# cancelSchedule(params, securityContext) → {Promise.<Object>}

Parameters:

# checkTranscriptions(params, securityContext) → {Promise.<Object>}

Parameters:

# cleanParticipantName()

# createDbAdapter()

# createMCPServer(configopt) → {Object}

Parameters:

# createRoom(params, securityContext) → {Promise.<Object>}

Parameters:

# createSchedule(params, securityContext) → {Promise.<Object>}

Parameters:

# ensureAIClient()

# async executeAgent()

# async executeDirect()

# async executeSimple()

# expandGraphNode(params, securityContext) → {Promise.<Object>}

Parameters:

# extractDisplayNameFromEmail()

# extractInfo(params) → {Promise.<Object>}

Parameters:

# filterOutBots()

# findCollaborators(params, securityContext) → {Promise.<Object>}

Parameters:

# findRecordingsByTopic(params, securityContext) → {Promise.<Object>}

Parameters:

# findRelatedTopics(params, securityContext) → {Promise.<Object>}

Parameters:

# findSimilarTopics(params, securityContext) → {Promise.<Object>}

Parameters:

# async findStoredRoom(roomName, cleanedOwner, apiKey) → {Promise.<(Object|null)>}

Parameters:

# formatDate()

# formatDuration()

# formatExamplesForPrompt()

# formatMeetingDate()

# formatStoredSummary()

# formatTranscriptWithSentiment()

# generateResponse(params) → {Promise.<Object>}