Direct Answer
Connecting Sparx EA to Salesforce Agentforce via MCP involves six steps: confirming prerequisites (EA GraphLink deployed with MCP Server active, Agentforce Studio access), registering the EA MCP Server as an Agentforce external data source, configuring the Agentforce agent topic for EA queries, testing with representative sample queries, building Agentforce actions from EA data, and optionally connecting to MuleSoft for broader data integration. The MCP mechanism is the same technical standard as the Microsoft Copilot connection: the difference is the registration surface (Agentforce Studio rather than Copilot Studio) and the user surface (Salesforce CRM and Service Cloud rather than Teams and Outlook). For Salesforce-first organizations, this is the path to EA intelligence embedded in the tools your customer-facing and operations teams already use.
Key Takeaways
- Agentforce connects to the EA MCP Server via the same MCP standard as Copilot: one MCP Server serves both.
- Registration is in Agentforce Studio (Salesforce admin environment), not Microsoft admin tools.
- Agentforce’s distinctive capability: agentic actions: not just answering questions, but triggering Salesforce tasks and workflows based on EA data.
- MuleSoft is the Salesforce data integration companion: equivalent to Microsoft Fabric in the Microsoft stack.
- Prerequisites: EA GraphLink Interface B (MCP Server) active and network-accessible from Agentforce.
- MDG quality gate applies: answer quality is bounded by repository governance quality.
- Sparx Services delivers this integration through the Connect engagement for Salesforce-ecosystem clients.
Prerequisites
Before beginning, confirm the following are in place:
EA GraphLink Interface B (MCP Server) is deployed:
- MCP Server is active and the endpoint is accessible (HTTPS endpoint confirmed)
- Authentication is configured (API key or OAuth 2.0 available for Agentforce integration)
- The MCP Server capability declaration has been verified: check that the element types and query patterns relevant to your Agentforce use cases are included
- Test the MCP Server endpoint independently (curl or Postman) to confirm it responds correctly before attempting Agentforce registration
Salesforce Agentforce access:
- Agentforce Studio is available in your Salesforce org (requires Agentforce license)
- You have System Administrator or Agentforce Admin permissions in the Salesforce org
- Your Salesforce org is on a release that supports external MCP data sources (Summer ’25 or later)
Network connectivity:
- Agentforce must be able to reach the EA MCP Server endpoint
- If the MCP Server is within your corporate network, a Salesforce Connected App or a secure outbound gateway is needed to allow Agentforce cloud-to-private-network connectivity
- Alternatively, if EA GraphLink is hosted in a cloud environment with a public (authenticated) endpoint, direct connectivity is simpler
Scoping: Before configuration, agree on the EA query scope for Agentforce. What questions will Agentforce answer from the EA repository? Common Agentforce-specific use cases include: “Which systems support the customer portal used by this account?”, “What is the architecture of our Customer 360 platform?”, “Which applications are reaching end-of-life that affect customer service operations?” Define these use cases: they shape the agent topic configuration.
Step 1: Register the EA MCP Server as an Agentforce External Data Source
In Salesforce Setup → Agentforce → External Data Sources:
- Navigate to Setup → Agentforce → External Data Sources
- Click “New External Data Source”
- Configure the source:
– Label: Enterprise Architecture Repository – API Name: EA_Repository (auto-generated) – Type: MCP (Model Context Protocol) – Server URL: [Your EA GraphLink Interface B endpoint URL] – Authentication Method: API Key – API Key: [The key generated in EA GraphLink configuration for the Agentforce integration]
- Save the external data source
- Click “Test Connection”: confirm that Agentforce can reach the MCP Server and receive a capability declaration response
Troubleshooting connectivity: If the test connection fails:
- Verify the MCP Server URL is correct and the server is running
- Check that Agentforce’s IP ranges are allowed through your network firewall for outbound connections to the MCP Server
- Verify the API key is correctly entered (no leading/trailing spaces)
- Check EA GraphLink logs for connection attempt records
Step 2: Configure Agentforce Agent Topic for EA Queries
An Agentforce agent topic defines what an agent can do and talk about. Create a topic specifically for EA repository queries:
In Agentforce Studio → Agents → [Your Service Agent] → Topics → New Topic:
Topic configuration:
Topic Label: Enterprise Architecture Intelligence
Topic Description (the system prompt for this topic): > This topic allows the agent to answer questions about the organization’s enterprise architecture, including information about business applications, technology platforms, business capabilities, system dependencies, and architecture decisions. The agent should query the Enterprise Architecture Repository for specific, current data rather than relying on general knowledge. Use this topic when customers or internal users ask about which systems support specific business functions, technology lifecycle information, or architectural context for service delivery.
Instructions (guidance for the AI on how to handle queries): > Always query the EA Repository external data source for specific EA questions. Do not answer EA questions from general knowledge alone: the repository contains the organization’s specific architecture data. When asked about applications, capabilities, or architecture decisions, retrieve current data from the repository. If the repository returns no results for a query, inform the user that data may not be available for that topic and suggest contacting the enterprise architecture team.
Actions (what the agent can do in this topic):
- Query EA Repository (via the registered MCP external data source)
- Draft follow-up summary for the user
- (Optional) Create Salesforce task for architecture review (Step 5)
Step 3: Test with Sample Queries
With the external data source registered and the agent topic configured, test with representative queries in Agentforce Studio’s preview mode:
Test 1: Application capability query: > “Which applications support our Customer Onboarding capability?”
Expected: Agentforce queries the EA MCP Server for Realisation relationships where the target is the “Customer Onboarding” Capability element, returns Application Component names.
Test 2: Lifecycle status query: > “Are any of our customer-facing applications reaching end-of-life in the next 12 months?”
Expected: Agentforce queries Application Components with End-of-Life lifecycle status, filtered to customer-facing elements (by business domain or capability linkage). Returns application names and EOL dates.
Test 3: Dependency query: > “What systems does our CRM platform depend on?”
Expected: Agentforce queries relationships from the CRM Application Component in the EA repository, returns connected technology nodes and application dependencies.
Test 4: Decision retrieval: > “What are our architecture standards for cloud infrastructure?”
Expected: Agentforce queries ArchitectureDecision elements with cloud-related keywords in their documentation, returns relevant accepted decisions.
Evaluating results: Do the answers reflect your actual repository data? Are element names consistent with what your architects have modeled? Discrepancies may indicate MDG governance gaps (elements not using the expected stereotype or named inconsistently) or query configuration issues.
Step 4: Configure Agentforce Topic for Production Use
After testing, refine the topic configuration based on test findings:
Scope refinement: If test queries returned too much data (all 400 applications when a more filtered answer was appropriate), add instructions to the topic guidance to scope queries more narrowly. Example: “When asked about customer-facing systems, filter to applications with business domain tagged as ‘Customer Management’ or that realize capabilities in the Customer Management domain.”
Exclusion rules: If the EA repository contains sensitive content (security architecture, system vulnerabilities) that should not be surfaced in the customer service context, add exclusion instructions: “Do not retrieve or discuss content from the Security Architecture package of the EA repository.”
Clarification handling: Add instructions for how the agent should handle ambiguous queries: “If a query could refer to multiple systems (e.g., ‘the CRM system’ when there are multiple CRM-related applications), ask the user for clarification before querying.”
Language adjustment: Configure the response format to match the context: customer-facing Agentforce agents should use plain language; internal technical agents can use more precise architectural terminology.
Step 5: Build Agentforce Actions from EA Data
This is Agentforce’s distinctive capability beyond simple querying: using EA data as input to automated Salesforce workflows.
Action 1: Create Architecture Review Task
When an internal user identifies an architecture concern (e.g., “the system our customer is using is on our decommission list”), Agentforce can automatically create a Salesforce task for the EA team:
In Agentforce Studio, create a new Flow-based action:
- Trigger: Agent identifies an application with End-of-Life status in a customer service context
- Action: Create a Salesforce Task (Assigned To: EA Team Queue, Subject: “Architecture Review Required: [Application Name]”, Description: auto-populated from EA data, Related To: the relevant Account or Case)
Action 2: Flag Architecture Risk on Salesforce Case
When a service case involves a system that the EA repository flags as high risk (End-of-Life, no migration plan, or security classification concern), update the Salesforce Case with an architecture risk indicator:
- Trigger: EA MCP query returns an application with high-risk attributes
- Action: Update Salesforce Case custom field
ArchitectureRiskFlag__c = True; add a Case Comment with the risk details from EA data
Action 3: Notify Architecture Team
When a recurring customer complaint pattern involves a specific capability (e.g., repeated complaints about the Customer Onboarding experience), Agentforce can notify the responsible EA team member based on capability ownership data from the EA repository:
- Trigger: Threshold of complaints related to a specific capability reached
- Action: Query EA MCP Server for the Business Owner tagged value on the Capability element; send Salesforce Chatter notification to the owner
Step 6: Connect to MuleSoft for Broader Data Integration
MuleSoft (Salesforce’s integration platform) is the Salesforce-stack equivalent of Microsoft Fabric: it enables EA data to be joined with other enterprise data sources for richer intelligence.
MuleSoft integration pattern:
- Deploy a MuleSoft API that queries EA GraphLink Interface A (GraphQL) on a schedule and writes EA data to a MuleSoft-managed data store
- From the MuleSoft data store, join EA data with Salesforce CRM data (accounts, products, service cases) and other connected systems
- Expose the enriched dataset via a MuleSoft Experience API consumed by Tableau for dashboards
- Optionally: expose the enriched dataset back to Agentforce as a secondary context source
Example enrichment queries in MuleSoft:
- “Which applications support products sold to our top 20 accounts?” (joins EA Application-to-Capability data with Salesforce Account-to-Product data)
- “Which customer-facing capabilities have End-of-Life applications?” (correlates EA lifecycle data with customer impact data)
MuleSoft integration is a more complex workstream than the direct MCP connection: it is appropriate for organizations where EA data needs to be combined with Salesforce CRM data for business intelligence, not just for AI querying.
Frequently Asked Questions
Q: Can Agentforce and Microsoft Copilot both connect to the same EA MCP Server simultaneously? Yes. The EA MCP Server supports concurrent connections from multiple MCP clients. Agentforce and Copilot can both be registered as MCP clients and query the same EA repository simultaneously. For organizations using both Salesforce and Microsoft 365, this enables EA intelligence in both ecosystems from a single EA GraphLink deployment. There is no conflict or competition between the two connections.
Q: Does Agentforce need to be configured per-agent, or is one EA MCP connection available to all agents? The external data source (MCP Server registration) is configured once at the org level. Individual agents access the external data source by referencing it in their topic configuration. Multiple Agentforce agents can use the same EA MCP external data source: a customer service agent, an internal support agent, and an architecture review agent can all query the same EA repository with different topic scopes and instructions.
Q: What Salesforce release is required for MCP external data source support? MCP external data source support in Agentforce became available from Summer ’25 (Salesforce API version 61.0). If your Salesforce org is on an earlier release, upgrade to Summer ’25 or later before attempting EA MCP Server registration. Check your Salesforce org’s current API version in Setup → Company Information.
Q: Can the Agentforce EA integration work with Salesforce Health Cloud or Financial Services Cloud? Yes. The Agentforce EA integration is at the platform level: it is available in any Salesforce cloud (Sales Cloud, Service Cloud, Health Cloud, Financial Services Cloud) that has Agentforce enabled. The industry-specific clouds add domain-relevant data objects and processes; the EA MCP connection adds architecture intelligence to whichever cloud context you are working in. Configuration steps are the same regardless of which Salesforce cloud you are using.
Q: How is authentication managed between Agentforce (cloud) and an on-premises EA MCP Server? For an on-premises or private-cloud EA MCP Server, a secure outbound gateway is needed to allow Agentforce (which runs in Salesforce’s cloud) to reach the private MCP Server. Options include: a Salesforce Private Connect configuration (using AWS or Azure private link), a secure reverse proxy exposing the MCP endpoint with strict IP allowlisting of Salesforce’s egress IP ranges, or deploying EA GraphLink in a cloud environment (rather than on-premises) so it is network-accessible to Agentforce. Sparx Services addresses the network architecture during the Connect engagement.
Q: Can we limit which EA data Agentforce can access? Yes. EA GraphLink Interface B configuration supports access controls that restrict which packages, element types, or tagged value dimensions are exposed via the MCP Server. For Agentforce-specific access, configure a dedicated API key that has read access only to the packages and element types relevant to the Agentforce use cases (e.g., Application Portfolio and Business Capabilities packages only: not Security Architecture or Financial Systems packages). This limits the scope of EA data accessible from the Salesforce context.
Ready to Connect Sparx EA to Salesforce Agentforce?
Sparx Services’ Connect engagement for Salesforce-ecosystem clients covers EA GraphLink deployment, Agentforce external data source registration, agent topic configuration, sample query testing, and initial agentic action design.
For Salesforce-first organizations, this is the path to EA intelligence embedded in CRM, service operations, and Agentforce workflows.
