Rethinking Complex Task Automation With Akvelon’s Agentic MCP Platform
Even the most advanced AI assistants lose focus when tasks get complex. They struggle to keep long context, coordinate multiple steps, and work reliably with massive or messy data and brownfield systems. That’s why we built Akvelon’s Agentic MCP Platform – a vendor-agnostic framework that lets us compose single-purpose sub-agents and orchestrate long-context workflows into one production-ready result.
Using this platform, we built a custom MCP for a global client and applied it to a ~30GB+ legacy repository with 100k+ artifacts and minimal documentation. On a representative, production-sized collection of artifacts (sub-project), initial repository review and analysis time dropped from ~4 workdays to ~1 work hour. The output was tidy documentation and a clear picture of the system.
Main capabilities of Akvelon’s Agentic MCP Platform:
- Compose single-purpose sub-agents for multi-step tasks
- Manage long context without cross-talk or intent drift
- Build composable workflows and reusable blocks
- Operate as an MCP consumer — connect to third-party MCP servers (e.g., Azure DevOps, GitHub, MS Learn) to use their tools/data within workflows
- Operate as an MCP producer – run our own MCP server to expose workflows/tools to any MCP-capable client (IDE/chat)
Why This Works
Instead of one model trying to do everything, our approach breaks work into single-goal steps. Each sub-agent focuses on one task, then hands control back. This keeps context clean, avoids mixed intents and repetition, and lets the orchestrator combine results into one coherent artifact.
How It Worked for a Global Technology Company
For a global technology company with a ~30GB+ legacy repository containing 100k+ artifacts and sparse documentation, we built a custom MCP using Akvelon’s Platform. The solution analyzed collections of artifacts (sub-projects) in parallel, summarized findings, generated structured documentation, and merged results into one consistent report.
Below is the client-specific MCP built with Akvelon’s platform:
A Bootstrapper creates the plan, an Orchestrator assigns single-goal tasks to Runners (agents using MCP tools like Azure DevOps, GitHub, and MS Learn) and composes one validated output.
What you’re seeing:
- The workflow is split into single-goal steps rather than one model doing everything.
- Runners use the right tool and context for each task, then hand control back.
- The Aggregator merges results into one coherent artifact (collection-level docs + repository-wide report).
Results for Our Client
- ~4 workdays → ~1 work hour per representative collection of artifacts (sub-project, initial analysis)
- Deliverable: structured, searchable documentation and a clear system view for refactoring and onboarding
- Qualitative impact: faster ramp-up and fewer repeats of manual review
This result reflects the specific implementation for this client. Outcomes may vary based on the environment, system size, quality, and goals. However, this approach consistently shortens cycles and clarifies legacy systems.
Technologies & Integrations
Languages & Runtime: TypeScript, Node.js;
Schema & Validation: Zod, Ajv;
Agentic & MCP: MCP TypeScript SDK, OpenAI Agents (JavaScript);
MCP Servers (consumed): Azure DevOps MCP Server, Azure MCP Server, GitHub MCP Server, Microsoft Learn MCP Server.
Where the Platform Helps Most
These are common scenarios where Akvelon’s Agentic MCP Platform + Framework deliver the greatest value.
Ready to Apply MCP?
We use our Platform to build custom MCPs that make long-context and brownfield work clear and fast. We’ll map your use cases and outline fit, approach, and integration.