Enterprise Architecture

What Is the Sparx EA MCP Server? Technical Guide for AI Integration

By Ryan Schmierer  ·  February 12, 2026

Published: 2026-04-18 Category: What Is Offering relevance: Connect

Direct Answer

The Sparx EA MCP Server is a native MCP-compliant server built into EA GraphLink that exposes the Sparx EA repository to any MCP-compatible AI tool. It is EA GraphLink Interface B. When the MCP Server is deployed, AI assistants including Claude, Microsoft Copilot, Google Gemini, Salesforce Agentforce, and ChatGPT Enterprise can discover the EA repository as a context source, query its elements, relationships, tagged values, and documentation, and generate natural-language answers about the organization’s architecture. This is not a custom API wrapper or a manual export: it is a standardized, governed, real-time connection between the EA repository and the AI assistant layer. EA GraphLink is licensed from Sparx Systems; Sparx Services implements and configures it through the Connect engagement. The MDG Technology governance quality of the repository determines the quality of answers the AI assistants return.

Key Takeaways

EA GraphLink: The Two-Interface Architecture

EA GraphLink provides two interfaces from the same Sparx EA repository:

Interface A: GraphQL for BI Tools: A GraphQL API endpoint that exposes the EA repository as a queryable data source. Power BI connects to this endpoint via the GraphQL connector. Tableau connects via the Web Data Connector. Any GraphQL-compatible analytics tool can query the EA data schema and retrieve elements, relationships, and tagged values for dashboard building and data analysis.

Interface B: MCP Server for AI Tools: An MCP-compliant server that exposes the EA repository as an AI context source. Any MCP-compatible AI assistant can register this server, discover its capabilities, and query it using the Model Context Protocol. This is the Sparx EA MCP Server.

Both interfaces read from the same underlying Sparx EA repository. The GraphQL interface is optimized for structured query and data retrieval (BI use case). The MCP Server is optimized for AI context provision: a richer capability model that includes resources, tools, and prompt templates appropriate for AI assistant integration.

What the MCP Server Exposes

The Sparx EA MCP Server makes the following repository content available to connected AI assistants:

Elements by type: Every element in the repository is accessible as its MDG-defined type. ArchiMate element types: Capability, Business Process, Application Component, Technology Node, Data Object, Principle, Requirement, and all others: are queryable as first-class typed objects. AI assistants can filter: “give me all Application Components in the Finance domain with lifecycle status End-of-Life.”

The MDG-awareness is critical: elements are not returned as generic UML objects with a stereotype string that the AI must parse: they are returned as semantically typed entities with their attributes and relationships already structured.

Relationships: ArchiMate relationships (Realisation, Assignment, Association, Serving, Influence, Composition, Aggregation, and all others) are queryable with their source and target elements. An AI can answer “which Application Components realize the Customer Onboarding capability?” by querying Realisation relationships where the target is the Customer Onboarding Capability element.

Tagged values: All custom attributes defined by MDG Technology extensions are queryable as structured dimensions. If an Application Component has a tagged value Lifecycle Status with value Active, the AI can filter on Lifecycle Status = Active. The enumeration values defined in the MDG (Active, Tolerate, Migrate, Eliminate, End-of-Life, Decommissioned) are the structured vocabulary the AI uses.

Package structure: The hierarchical organization of the repository: which packages are at the top level, which elements belong to which packages: is navigable. AI assistants can scope queries to specific packages: “in the Business Architecture package, which Business Processes have no documented owner?”

Documentation: Element notes and documentation fields are included in MCP Server responses. This enables the AI to answer questions that require reading textual context: not just structured attribute queries. Architecture Decision Record documentation (the Context, Decision, Consequences sections) is available to AI assistants for precedent surfacing.

Diagram metadata: Diagram names and the elements that appear in them are accessible. The AI cannot see rendered diagram images: it sees diagram metadata. This enables queries like “which diagrams in the Application Architecture domain show the CRM components?”

How MCP Discovery Works

MCP’s discovery mechanism is what distinguishes it from traditional API integration. When an AI assistant first connects to an MCP Server:

  1. Capability declaration: The MCP Server sends a structured description of what it offers: the element types available, the query patterns supported, the tools and resources accessible. This is a machine-readable schema that the AI can reason about.
  1. AI understanding: The AI assistant parses the capability declaration and understands what the EA MCP Server contains. It does not need to be told “call endpoint X with parameter Y to get Application Components”: it knows from the declaration that Application Components are available and what attributes they have.
  1. Contextual querying: When a user asks a question that the AI determines requires EA repository data, the AI formulates an appropriate query against the MCP Server based on its understanding of the server’s capabilities. The query is contextually generated: not pre-programmed.
  1. Response integration: The AI receives structured data from the MCP Server and synthesizes it into a natural-language answer for the user. The user sees a fluent response, not raw data.

This discovery-and-query model means that as the EA repository evolves: new element types added via MDG updates, new tagged value dimensions defined: the AI’s understanding updates accordingly. The MCP Server re-describes its capabilities on connection; the AI adapts.

The MDG Quality Gate

This is the most important technical concept for EA teams deploying the Sparx EA MCP Server: the quality of AI answers is bounded by the quality of MDG governance in the repository.

Here is why:

The MCP Server serves data from the repository as it exists. If ArchiMate elements are correctly stereotyped (Application Component is an Application Component, not a generic Class with a note saying it is an application), the AI receives precise typed data. If elements use the correct tagged value vocabulary (Lifecycle Status = Active / End-of-Life / Decommissioned: governed enumeration values), the AI can filter and categorize accurately.

If instead the repository contains generic UML elements with informal stereotypes, ad-hoc tagged values (someone typed “EOL” instead of “End-of-Life”), or empty mandatory fields (no owner recorded), the AI receives imprecise, incomplete data. Its answers reflect these gaps.

Example: A well-governed repository with consistent Lifecycle Status tagged values across all Application Component elements allows the AI to answer: “How many of our applications are in End-of-Life status?” with a precise number and a list.

A repository where lifecycle status is recorded as a free-text note in element documentation, without a structured tagged value, cannot answer this query precisely: the AI would need to parse free text from thousands of element notes, and the answer quality would be poor.

This is why Sparx Services treats MDG Technology governance (Amplify engagement) as a prerequisite or parallel workstream to EA GraphLink deployment (Connect engagement). The Amplify work: defining the MDG stereotypes, tagged value enumerations, validation rules, and naming conventions: is the work that makes the MCP Server produce trustworthy AI intelligence.

Which AI Tools Can Connect

As of 2026, MCP-compatible AI tools that can connect to the Sparx EA MCP Server include:

The EA MCP Server is not locked to any of these: it is vendor-neutral. As new MCP-compatible tools emerge, they can connect without changes to the MCP Server configuration.

Licensing Model

EA GraphLink (which includes both Interface A and Interface B/MCP Server) is licensed by Sparx Systems. The license is typically per-repository or based on usage tier. Sparx Services, as a Sparx EA partner, advises on EA GraphLink licensing and coordinates procurement.

Sparx Services implements EA GraphLink through the Connect engagement. This includes:

The implementation is a one-time engagement. Once deployed, the MCP Server runs continuously: it does not require per-query configuration or per-AI-tool custom development.

Frequently Asked Questions

Q: Is the Sparx EA MCP Server the same as the Prolaborate stakeholder portal? No. Prolaborate is a separate product that publishes EA repository content as browsable web views for non-architect stakeholders. The Sparx EA MCP Server (EA GraphLink Interface B) is an AI integration layer that enables AI assistants to dynamically query the repository. Prolaborate shows pre-published content in a browser; the MCP Server serves dynamic query responses to AI assistants. They solve different problems and are often deployed alongside each other.

Q: Does the MCP Server require a Sparx EA license per AI user? No. The Sparx EA MCP Server serves AI assistants through the EA GraphLink integration: AI assistants are not Sparx EA users. Sparx EA licenses are required for architects who model in the Sparx EA client. The AI assistants querying the MCP Server do not consume Sparx EA licenses. EA GraphLink licensing (separate from Sparx EA modeling licenses) covers the API and MCP Server access.

Q: Can the MCP Server be restricted to query only specific packages or element types? Yes. The EA GraphLink Interface B configuration supports access controls that can limit which packages, element types, or tagged value dimensions are exposed via the MCP Server. This is useful when the repository contains sensitive content that should not be accessible to all AI assistants (e.g., HR system details, security architecture). Access restrictions are configured during the Connect engagement based on the organization’s data governance requirements.

Q: What is the query performance of the MCP Server for large repositories? EA GraphLink includes caching and query optimisation for both Interface A and Interface B. For typical AI assistant queries (which return element lists, relationships, or metadata for a focused domain), response times are in the range of 1–5 seconds for repositories of tens of thousands of elements. broad queries (return all elements of all types) may take longer and are typically addressed through focused query scoping rather than broad retrieval. Sparx Services tunes query performance during the Connect engagement.

Q: How is the MCP Server authentication managed? The EA GraphLink Interface B supports API key authentication and OAuth 2.0 for enterprise authentication integration. AI assistants connecting to the MCP Server are authenticated before they receive any repository access. API keys are managed in the EA GraphLink configuration; OAuth integration can connect to enterprise identity providers (Azure AD, Okta). Authentication configuration is set up during the Connect engagement.

Q: Does the MCP Server work with on-premises Sparx EA installations? Yes. EA GraphLink Interface B can be deployed alongside an on-premises Sparx EA and Pro Cloud Server installation. The MCP Server endpoint can be exposed within the corporate network (accessible to internal AI tool deployments) or through a secure gateway (for cloud-hosted AI tools). Network architecture for MCP Server access is designed during the Connect engagement, balancing accessibility with security requirements.

Q: What happens when new elements are added to the Sparx EA repository? The MCP Server reads from the live repository. New elements added by architects are immediately available for AI querying: there is no synchronisation step, no export, and no cache invalidation required. The AI assistant’s next query will include the new elements. This real-time data currency is one of the primary advantages of the MCP Server approach over manual export or scheduled data synchronisation.

Q: Can the Sparx EA MCP Server be used to write to the repository? No. The current implementation of EA GraphLink Interface B is read-only. AI assistants can query and retrieve repository data but cannot create, modify, or delete elements, relationships, or tagged values via the MCP Server. Write-back capability is under development for future EA GraphLink releases. For now, the MCP Server is the intelligence query layer; all repository modifications are made by architects through the Sparx EA client.

Ready to Deploy the Sparx EA MCP Server?

Sparx Services’ Connect engagement covers EA GraphLink procurement coordination, full installation and configuration of both interfaces, AI assistant registration, MDG quality assessment, and initial query validation.

Your EA repository can be AI-queryable within a Connect engagement: with every MCP-compatible AI tool in your organization drawing on governed architecture intelligence.

Talk to Sparx Services about deploying the EA MCP Server →

Share this article
LinkedIn

Related insights

More on Enterprise Architecture

Ready to make your EA investment work harder?

Talk to a Sparx Services architect about where your organization is on the journey and what the next stage looks like.

Talk to us See the journey