EA GraphLink’s MCP Server connects your Sparx EA repository to Claude (Anthropic). This is the MCP interface path: the same protocol interface used by Copilot and Agentforce integrations, but with a fundamentally different purpose.
Claude is a universal MCP-compatible AI. It has no Microsoft or Salesforce ecosystem dependency. It is not embedded in a business platform. It does not bring architecture data to the widest possible organizational audience. What it does instead is different and complementary: it reasons deeply on complex, multi-layered problems: exactly the kind of problems enterprise architecture involves.
EA GraphLink’s MCP Server gives Claude live access to your repository. With that context, Claude can analyze architectural trade-offs, synthesize large model sections into coherent narratives, generate documentation that reads like a architect wrote it, and work through “what does this mean for our architecture?” questions grounded in actual repository data rather than general knowledge.
The MDG Technology definition governs what Claude can access. EA GraphLink transforms the physical repository schema using the MDG Technology definition. Poor MDG quality means poor context for Claude’s reasoning. MDG readiness is assessed in Discover and established in Deploy. Claude’s output quality depends directly on the quality of what EA GraphLink exposes.
Live EA repository context for genuine architectural reasoning: not retrieval and display, but analysis and synthesis.
Your current state: An architect is preparing for a governance board review. She has three hours to turn a complex section of the repository into a coherent narrative that explains the current state, the gaps, and the recommended path forward. She spends most of those three hours reading elements, pulling dependencies, and writing connecting prose. The analysis is good. The process is slow.
Your future state: Claude has access to the same repository section via EA GraphLink. The architect asks Claude to synthesize the relevant ArchiMate domain into a briefing: summarize the current state, identify gaps against the target architecture, and flag the decisions that need governance attention. Claude reads the repository, synthesizes the information, and produces a structured draft. The architect reviews, corrects, enriches, and arrives at the governance board with a better briefing in a third of the time.
This is not search-and-retrieve. Claude analyzes, interprets, identifies patterns, and reasons across the model. The value is not that it can display what’s in the repository: it’s that it can think with it.
Copilot and Agentforce are optimized for reach. They bring architecture data to the widest possible audience in the tools those people already use. Claude is optimized for depth. It brings genuine reasoning capability to complex architectural problems where surface-level answers are not sufficient.
These are not competing use cases. An organization can deploy both: Copilot or Agentforce for broad stakeholder access, Claude for architect-level analysis. The EA GraphLink MCP Server supports both simultaneously.
Claude is a thinking partner for the problems that require thinking. Analysis preparation. Documentation drafting. Gap identification. Cross-layer reasoning. Coaching junior architects against real repository content. These are problems where depth matters, where generic AI responses are not useful, and where live repository context makes the difference.
Documentation generation at scale. Review preparation. Governance board briefings. Cross-cutting gap analysis. Synthesis tasks that currently consume architect time become Claude-assisted workflows that produce drafts the architect reviews rather than writes from scratch.
Architecture Review Preparation
A governance cycle is approaching. An architect needs to brief the board on the current state of the application landscape in a specific business domain. She asks Claude to synthesize the relevant section of the repository into a briefing: current applications, their business capability alignment, technology health, open risks, and the recommended decisions for the board. Claude reads the repository via EA GraphLink, produces a structured draft. The architect reviews and refines. Board briefing preparation time is reduced substantially.
Cross-Layer Trade-Off Analysis
A technical decision is under review: should a critical application be migrated to the cloud? The architect asks Claude: “Based on our current dependency map in the EA repository, what are the architectural implications of migrating this application to Azure? What depends on it, what integration patterns would need to change, and what’s the highest-risk element of the transition?” Claude traces dependencies in the repository, reasons across the model, and produces a structured trade-off analysis grounded in actual repository data: not hypothetical cloud migration advice.
Documentation Generation from Repository Content
The architecture practice needs to produce element descriptions, view narratives, and stakeholder communications. Currently, junior architects write these from scratch or skip them due to time pressure. Claude with EA GraphLink access can generate first drafts for element descriptions, view-level narratives, and domain summaries that read like a architect wrote them: because they’re grounded in actual repository content. Senior architects review and approve rather than write from nothing.
Gap Analysis Against Standards or Target Architecture
An architect needs to identify which elements in a given domain have no assigned owner, no documented rationale, or no relationship to the target architecture. Claude with repository access can run structured queries against the model: “Which ArchiMate application components in the customer domain have no assigned owner and no relationship to a business capability?” The gap report surfaces in minutes rather than hours of manual model review.
EA Coaching for Junior Architects
A junior architect is working on a section of the repository. She doesn’t fully understand why a particular pattern was chosen. She asks Claude (with repository access): “Why is this integration pattern used in our payment domain? What are its trade-offs compared to alternatives?” Claude can explain the pattern in context: grounded in what the repository actually shows: and provide the contextual learning that helps the junior architect build judgment, not just produce model content.
| Role | Primary Benefit | Secondary Benefit |
|---|---|---|
| EA Architect | Deep analysis partner for complex architectural problems | Documentation generation, review prep, gap identification |
| Architecture Manager | Board briefings and governance documentation at scale | Cross-cutting gap analysis without manual model review |
| Technical Architect | Architecture-aware reasoning during design work | Trade-off analysis grounded in live dependency data |
| MBSE Architect | Synthesis and documentation of complex model sections | Gap identification against requirements or standards |
| Junior Architect | Contextual coaching against real repository content | Accelerated understanding of existing patterns and rationale |
| Business Stakeholder | Low value: Copilot or Agentforce serve them better | (These personas should use ecosystem-embedded tools) |
Claude is optimized for depth: not for ecosystem reach.
Where Copilot and Agentforce bring architecture data to the widest possible audience in existing tools, Claude brings genuine reasoning capability to complex architectural problems. The two serve different needs. They are complementary, not competing.
No ecosystem dependency. Claude doesn’t require M365 licensing, Salesforce licensing, or any platform commitment. An organization on any combination of tools can add Claude to an EA GraphLink deployment. There is no platform infrastructure to acquire: only an Anthropic API key and configuration.
The quality of AI reasoning scales with context quality. Claude with live EA repository context is substantially more useful than Claude with general knowledge. The difference between “here’s what architectural patterns generally look like for this problem” and “here’s what your actual dependency map shows and why that matters for this decision” is the entire value of the integration. EA GraphLink provides that live context.
Pay-per-token pricing means cost scales with value. Unlike Copilot or Agentforce, which require per-seat licensing regardless of usage, Claude API pricing is usage-based. Small teams of architects using Claude intensively can be cost-competitive with per-seat licensing models. Evaluate actual usage patterns.
Architecture data leaves the environment to Anthropic’s API. Claude is a cloud API. Repository data queried through EA GraphLink flows to Anthropic’s infrastructure to generate responses. For regulated industries: healthcare, financial services, government: this requires a compliance and data governance review before deployment. This is not a blocker for most organizations, but it is a governance conversation that must happen before rollout.
Not embedded in M365 or Salesforce: less reach to business stakeholders. Claude is a architect tool. It does not bring architecture data to business stakeholders in the tools they use for work. If reaching business stakeholders in Teams or Salesforce is the priority, Copilot or Agentforce is the right integration. Claude serves architects and technical architects who use it directly.
API cost scales with usage. Pay-per-token pricing is an advantage for low-volume, high-value use. It becomes expensive at high volume. Model the expected usage pattern: number of architects, query frequency, average query complexity: before committing to Claude API at scale.
Requires architects to actively change their workflow. Claude is not ambient. It doesn’t surface information when a stakeholder is in a meeting. Architects need to choose to use it, form their query, and review the output. The value is real, but adoption requires behavior change in the architecture practice.
For regulated industries: evaluate Azure OpenAI instead. If data residency or sovereignty requirements prevent sending data to Anthropic’s API, Azure OpenAI provides equivalent reasoning capability with data staying within an Azure tenant. The capabilities differ at the margin; the governance posture differs significantly. Regulated organizations should evaluate both options.
EA GraphLink with MCP Server enabled. The MCP interface is how Claude connects to the EA repository. Confirm your EA GraphLink deployment includes an active MCP Server endpoint with your Sparx Systems account manager.
Anthropic API access. An Anthropic API key is required. This is procured directly from Anthropic: Sparx Services provides a bill of materials, not resale. API access can be scoped to a team or individual architects during pilot; expanded later based on adoption.
MCP client configuration for Claude. Claude’s desktop application and API support MCP tool connections. The EA GraphLink MCP Server is configured as a tool available to Claude. Sparx Services handles EA GraphLink configuration during Connect.
MDG Technology quality sufficient to trust analytical output. Claude’s reasoning quality depends entirely on the quality of the repository context it receives. EA GraphLink transforms the physical repository schema using the MDG Technology definition. Incomplete or inconsistent MDG definitions produce incomplete or inconsistent context for Claude. MDG readiness is assessed in Discover and established in Deploy. Do not deploy Claude integration before validating MDG quality: the analytical output will be unreliable.
Architect workflow planning. Claude integration changes how architects work. Which workflows benefit? Review preparation? Documentation? Gap analysis? Define the priority use cases, plan the workflow changes, and set expectations for output review. Claude produces drafts and analysis for architect review: not final deliverables without oversight.
Architecture review preparation hours per governance cycle.
How many cycles does your architecture practice run per year? How many hours does preparation currently require per cycle? Claude-assisted synthesis typically reduces preparation time by 40–60%.
Documentation generation time.
How many elements in your repository lack descriptions, rationale, or view narratives? At what hourly rate do architects write this content? Claude-assisted drafting typically reduces documentation time by 50–70% (draft generation) with additional architect review time.
Formula:
Architecture review preparation:
Review cycles per year
× architect hours per review preparation cycle
× reduction from Claude-assisted synthesis (40–60%)
× architect hourly rate
= annual hours and cost reclaimed
Documentation:
Number of elements requiring documentation
× hours to document manually per element
× reduction from Claude-assisted drafting (50–70%)
× architect hourly rate
= documentation efficiency gain
Gap analysis:
Gap analysis cycles per year
× manual hours per gap analysis cycle
× reduction from Claude-assisted queries
× architect hourly rate
= gap analysis efficiency gain
Example:
4 governance cycles/year × 20 hours prep/cycle × 50% reduction × $150/hour = $6,000 reclaimed
500 undocumented elements × 0.5 hours per element × 60% reduction × $150/hour = $22,500 efficiency gain
6 gap analyzes/year × 8 hours each × 70% reduction × $150/hour = $5,040 efficiency gain
Total: ~$33,500 annual value (conservative estimate for mid-size practice)
Organizations running larger governance programs, larger repositories, or more documentation backlogs will see proportionally larger returns. The value compounds as the repository grows and governance frequency increases.
Azure OpenAI
Equivalent reasoning capability to Claude, with data staying within an Azure tenant rather than flowing to Anthropic’s API. The right choice for regulated industries with data residency requirements, or organizations already standardized on Azure AI services. Capabilities differ at the margin: Claude typically performs well on nuanced analytical tasks; Azure OpenAI on GPT-4 is comparable for most EA use cases. The decision is primarily a governance and ecosystem choice.
Kernaro AI Hub
Kernaro is Sparx Systems’ purpose-built architecture intelligence platform. It connects to the EA repository without data leaving your environment, and is specifically designed for architectural queries and intelligence. Pros: no data egress, architecture-specific design, Sparx-native. Cons: less general reasoning depth for complex multi-layer analytical problems, narrower use case range than Claude. Kernaro is excellent for structured architectural queries and portal-based access; Claude is stronger for open-ended reasoning and synthesis tasks.
Microsoft Copilot
Better ecosystem reach to M365 users: brings architecture data to the broadest possible stakeholder audience in Teams, Outlook, and Word. Less analytical depth for complex multi-layer architectural problems than Claude. The right choice when the priority is organizational reach; Claude is the right choice when the priority is depth of analysis for architects.
Q: What can Claude do with access to the EA repository?
A: Claude can query your repository via EA GraphLink, reason across the content, synthesize sections into narratives, identify patterns and gaps, analyze trade-offs between architectural options, generate documentation drafts grounded in repository content, and answer complex analytical questions about your architecture. It is not limited to retrieving and displaying information: it can reason with the information. The quality of its reasoning depends on the quality of the repository content it receives.
Q: Does using Claude mean my architecture data is used to train AI models?
A: Not by default. Anthropic’s current API terms do not use API traffic to train models without explicit opt-in. This is distinct from consumer-facing Claude products. Verify current Anthropic API data handling policies before deployment: policies can change, and regulated industries need this confirmed in writing. For organizations where this remains a concern regardless, Azure OpenAI provides equivalent capability with data staying in an Azure tenant.
Q: What is the difference between Claude and Kernaro AI Hub?
A: Kernaro is purpose-built for architecture intelligence: it is specifically designed to work with EA repositories and architectural data. It does not send data externally and is native to the Sparx ecosystem. Claude is a general-purpose reasoning model that gains architecture context through EA GraphLink’s MCP interface. Claude’s strength is depth of reasoning on complex, open-ended problems. Kernaro’s strength is architecture-specific design, no data egress, and native integration. For regulated industries or organizations where data leaving the environment is not acceptable, Kernaro is the right path. For organizations seeking deep reasoning on complex architectural problems without data residency constraints, Claude is a strong complement.
Q: Is Claude suitable for regulated industries?
A: It depends on the specific regulation and your organization’s interpretation of it. Claude is a cloud API: EA repository data flows to Anthropic’s infrastructure for query processing. For industries with strict data residency requirements (some healthcare, financial services, government contexts), this may require review or may be prohibited without additional safeguards. For many regulated industries, API-based cloud processing is acceptable with appropriate data handling agreements. Evaluate with your compliance team before deployment. Azure OpenAI is the alternative if data residency is a hard constraint.
Q: Do I need a separate Claude API key for each architect?
A: No. Claude API access is typically managed through a single organization API key with usage tracked centrally. Individual architects authenticate through your organization’s API key rather than requiring individual accounts. Access controls and usage monitoring can be configured at the organizational level. Consult Anthropic’s current API documentation for the latest access management options.
Q: How current is the architecture data Claude can access?
A: EA GraphLink queries the live EA repository at the time of each Claude request. The data Claude receives reflects the repository at query time: not a cached snapshot or periodic export. If an element was updated in the repository this morning, Claude’s response to a query this afternoon will reflect that update. Currency of data depends on how actively the repository is maintained, not on the integration’s refresh schedule.
Q: Can Claude write back to the EA repository?
A: No. EA GraphLink’s MCP Server is a read interface. Claude can query and reason over repository content but cannot modify elements, create new content, or update existing records. All output from Claude is external to the repository: narratives, analyzes, documentation drafts: and must be manually reviewed and entered into the repository if the architect chooses to incorporate them. This is the correct design: AI-assisted analysis reviewed by architects before it enters the authoritative record.
Q: What is the difference between Claude and Claude Code for EA users?
A: Claude (the AI assistant, accessed via API or desktop application) is the tool described in this guide: used by architects for analysis, documentation, and reasoning. Claude Code is a software development tool: an AI coding assistant embedded in terminal and IDE workflows. Claude Code can also connect to EA GraphLink’s MCP Server via the same protocol, but it is designed for technical architects doing development work: querying approved interfaces, checking dependency maps during code generation, validating architectural constraints. If your audience is software architects and engineers working in code, see the Cursor and Claude Code integration guide. If your audience is EA architects and architecture managers doing repository work, this guide describes the right tool.
Claude integration gives your architecture practice a genuine reasoning partner for the problems that require depth: review preparation, trade-off analysis, documentation, gap identification, and contextual coaching. It complements rather than replaces ecosystem-embedded tools like Copilot and Agentforce.
The journey starts with assessing your repository foundation: MDG quality, governance structure, and the specific workflows you want to improve. If you haven’t completed a Discover assessment, that’s the right first step.
If you’re ready to connect Claude to your EA repository:
[Start Your Connect Engagement]
Questions about which AI integration is right for your architecture practice?
[Let’s Discuss Your AI Strategy]
Talk to a Sparx Services architect about where your organization is on the journey and what the next stage looks like.