Tool Integrations
Connect your agents to the business systems, knowledge sources, and custom APIs your use case depends on, before the agents themselves are designed.
Control Hub connects your agents to your existing business systems, knowledge bases, and custom APIs. This page is the catalogue of what you can connect and what each connector does.
The Outcome
Every system your use case reads from or writes to is reachable by an agent, through a connector you have chosen deliberately.
Before You Start
Know which systems the use case actually touches. Agentic Process Transformation names the applications behind the work - an Agent as Tool pattern usually means an agent has to reach one of those systems directly.
Access is the slow part, so line it up early. Per-user OAuth connectors authenticate as an individual user, which means each user grants access with their own credentials and permissions. For a custom connector you need the MCP server endpoint URL plus an OAuth client ID and secret before you can add it at all.
Start Here
Write down the systems the use case reads from and writes to, then work down that list against the tables below: use a pre-configured MCP connector where one exists for the system, and a custom MCP connector where one does not.
Web Search
Enable agents to search the internet for real-time information. Multiple providers are available:
| Provider | Description |
|---|---|
| Playwright | Built-in web scraping, no configuration required |
| Tavily | AI-optimized search API |
| Brave | Privacy-focused search API |
MCP Connectors
Connect to your existing tools with pre-configured integrations:
| Category | Available Integrations |
|---|---|
| Communication | Slack, Microsoft Teams, Gmail |
| Project Management | Jira, Asana, Smartsheet |
| Documentation | Confluence, SharePoint, Google Drive |
| CRM | Salesforce, Zendesk, ServiceNow |
| Calendar | Google Calendar, Zoom |
| Storage | OneDrive, SharePoint |
| HR | BambooHR |
| Incidents | PagerDuty |
Knowledge Base (RAG)
Connect your company documents so agents can search and reference them during conversations.
Data Sources
| Source | What It Does |
|---|---|
| S3 Bucket | Index documents from your AWS storage |
| Web Crawler | Crawl and index your documentation sites |
| Confluence | Search your Confluence wikis |
| SharePoint | Access SharePoint documents |
| Salesforce | Query Salesforce knowledge articles |
Supported File Types
- Documents: PDF, DOCX, PPTX
- Spreadsheets: XLSX, XLS, CSV
- Text: TXT, MD, JSON
Per-User OAuth Connectors
These connectors authenticate per-user, allowing agents to act on behalf of individual users with their own credentials and permissions.
| Connector | Capabilities |
|---|---|
| Slack | List channels, search messages, post messages, reply to threads |
| Gmail | Search emails, read messages, create drafts, send emails |
| Google Calendar | List events, find free time, create/update/delete events |
| Google Drive | Search files, read content, create/share/move files |
| Microsoft Teams | List teams/channels, read messages, send messages |
Custom Connectors
Connect to any custom API or internal tool using the Model Context Protocol (MCP).
To add a custom connector, provide:
- Endpoint URL - The URL of your MCP server
- OAuth Client ID - Your OAuth application client ID
- OAuth Client Secret - Your OAuth application client secret
This allows you to extend your agents with any internal tools or third-party APIs that aren't covered by the pre-built integrations.
Knowledge Graph (GraphRAG)
The Graph sub-section in the Data tab lets you build a domain property knowledge graph from your uploaded documents. The graph enriches your agents with structured relationship data - going beyond keyword search to let agents reason over entity connections.
How It Works
- Navigate to the Data tab and select Graph
- Upload or select documents already in your Documents store
- The AI constructs a property graph: entities become nodes, relationships become edges, types and attributes are inferred from document content
- The graph is visualized live on a Cytoscape.js canvas - nodes are color-coded by type, edges are labeled, and the layout is interactive
- An editable Mermaid pipeline shows the extraction steps - you can adjust the pipeline before running
Node & Edge Configuration
| Element | What You Define |
|---|---|
| Node Types | Entity categories (Person, Organization, Product, Concept, etc.) |
| Edge Types | Relationship labels (works_at, references, depends_on, etc.) |
| Attributes | Per-node properties (id, name, description, metadata) |
Using the Graph in Agents
Once built, the knowledge graph is available as a tool for any agent in the Agentic System tab. Agents query it using natural language; the platform translates queries to graph traversals and returns structured context alongside the agent's retrieved documents.
GraphRAG is particularly effective for use cases with complex entity relationships: financial regulatory networks, healthcare ontologies, technical dependency graphs, and organizational hierarchies.
What Good Looks Like
- Every system the use case depends on is either covered by a pre-configured connector or has a custom MCP connector defined for it
- Connectors are in place because the use case needs them, not because they were on the list
- Where an agent acts on behalf of a person, a per-user OAuth connector is used so the agent inherits that user's permissions rather than a shared identity
- Knowledge Base sources index the documents where they already live - S3, Confluence, SharePoint, Salesforce, a crawled site - instead of duplicate copies that will drift
- If the use case needs live external information, one web search provider has been chosen on purpose: Playwright needs no configuration, Tavily and Brave are external search APIs
Where This Leads
- Governance & Guardrails - every connector widens what agents can reach, so set the access and compliance boundaries before any of this runs against production systems
- Data - the Connections sub-tab is where you test a connection and bind it to agents
- Agentic System - Launch, where these tools get assigned to individual agents