Skip to main content

​
Description

Agent Composer provides multiple ways to integrate with external systems, enabling agents to take actions, retrieve data, and interact with enterprise services.

​
Integration Options

IntegrationUse CaseStep Type
MCPConnect to MCP-compatible servicesMCPClientStep
WebhooksCall REST APIs and trigger actionsWebhookStep
Code ExecutionRun Python code for calculationsCodeExecutionStep
SalesforceQuery Salesforce dataSalesforceSOSLStep
Web SearchSearch the web for current informationWebSearchStep
Other AgentsQuery other Contextual AI agentsContextualAgentStep

​
Salesforce Integration

Use SalesforceSOSLStep to execute SOSL searches against Salesforce: 
salesforce_search:
  type: SalesforceSOSLStep
  config:
    client_id: "${SF_CLIENT_ID}"
    client_secret: "${SF_CLIENT_SECRET}"
    login_url: "https://login.salesforce.com"
    objects_to_return: "Case(Id, Subject, Status, Priority, Account.Name)"
    limit: 250
  input_mapping:
    query: __inputs__#query

​
Salesforce Configuration

ParameterTypeRequiredDescription
client_idstrYes*Salesforce connected app client ID
client_secretstrYes*Salesforce connected app client secret
login_urlstrNoSalesforce login URL (default: production)
objects_to_returnstrNoSOSL RETURNING clause
limitintNoMaximum results (default: 250)
timeoutintNoRequest timeout (default: 30)
*Alternative: Use username, password, and security_token for user-based auth. Use WebSearchStep for real-time web information: 
web_search:
  type: WebSearchStep
  config:
    model: "gemini-2.5-flash"
  input_mapping:
    query: __inputs__#query

​
Web Search Outputs

OutputTypeDescription
web_resultWebResultSearch results with sources

​
Query Other Agents

Use ContextualAgentStep to retrieve information from other Contextual AI agents: 
query_specialist:
  type: ContextualAgentStep
  config:
    agent_ids:
      - "agent-id-for-hr-docs"
      - "agent-id-for-legal-docs"
  input_mapping:
    query: __inputs__#query
This enables multi-agent architectures where specialized agents handle specific domains.

​
Vision Language Models

Use VisionLanguageModelStep for multimodal inputs: 
analyze_image:
  type: VisionLanguageModelStep
  config:
    model_id: "vertex_ai/gemini-2.0-flash"
    temperature: 0.0
    max_tokens: 4096
  input_mapping:
    query: __inputs__#query
    images: image_extractor#extracted_images

​
VLM Configuration

ParameterTypeDefaultDescription
model_idstr"vertex_ai/gemini-2.0-flash"Vision-capable model
temperaturefloat0.0Response creativity
max_tokensint4096Maximum response length

​
Example: Multi-System Agent

version: 0.1
inputs:
  query: str

outputs:
  response: str

nodes:
  create_message_history:
    type: CreateMessageHistoryStep
    input_mapping:
      query: __inputs__#query

  research:
    type: AgenticResearchStep
    ui_stream_types:
      retrievals: true
    config:
      tools_config:
        # Internal documentation
        - name: search_docs
          description: Search internal documentation and policies.
          step_config:
            type: SearchUnstructuredDataStep
            config:
              top_k: 50

        # CRM data via MCP
        - name: search_crm
          description: Look up customer records and history.
          step_config:
            type: MCPClientStep
            config:
              server_url: "https://crm.example.com/mcp"
              tool_name: "search"
              tool_args: '{"query": "$query"}'
              auth_headers:
                Authorization: "Bearer ${CRM_TOKEN}"

        # Salesforce cases
        - name: search_salesforce
          description: Find support cases in Salesforce.
          step_config:
            type: SalesforceSOSLStep
            config:
              client_id: "${SF_CLIENT_ID}"
              client_secret: "${SF_CLIENT_SECRET}"
              objects_to_return: "Case(Id, Subject, Status)"

        # Real-time web info
        - name: search_web
          description: Search the web for current information.
          step_config:
            type: WebSearchStep

      agent_config:
        agent_loop:
          num_turns: 10
          model_name_or_path: "vertex_ai/claude-sonnet-4-5@20250929"
          research_guidelines_prompt: |
            Choose the appropriate tool based on the query:
            - `search_docs` for policies and procedures
            - `search_crm` for customer information
            - `search_salesforce` for support cases
            - `search_web` for current events or external info

    input_mapping:
      message_history: create_message_history#message_history

  generate:
    type: GenerateFromResearchStep
    ui_stream_types:
      generation: true
    config:
      model_name_or_path: "vertex_ai/claude-sonnet-4-5@20250929"
    input_mapping:
      message_history: create_message_history#message_history
      research: research#research

  __outputs__:
    type: output
    input_mapping:
      response: generate#response