Workflows Overview

The Workflows section covers the automated pipelines that power MCP server creation and container image builds.

Generation Pipeline

Generating an MCP server from an API spec triggers the following pipeline:

OpenAPI Spec → LLM Code Generation → Kaniko Image Build → Kubernetes Deployment

Each stage is tracked as a generation job with real-time status and logs.

Stage 1: Code Generation

The platform sends your OpenAPI spec to an LLM (Ollama with a code model) along with a TypeScript MCP server template. The LLM produces:

• index.ts — MCP server implementation with typed tool handlers
• package.json — Node.js dependencies
• Dockerfile — multi-stage build for the server image
• tsconfig.json — TypeScript compiler config

The generated code is written to a persistent volume (/generated-code).

Stage 2: Image Build

Kaniko builds the generated Dockerfile inside a Kubernetes Job and pushes the resulting image to the in-cluster registry (ftn-ai-registry:5000).

Build logs stream in real time in the MCP server detail view.

Stage 3: Deployment

After a successful build, the platform deploys the MCP server as a Kubernetes Deployment in the project’s namespace and exposes it via a Service and Ingress.

Job Phases

Phase	Description
pending	Generation job queued
generating	LLM generating TypeScript server code
building	Kaniko building and pushing the Docker image
deploying	Kubernetes deploying the server
succeeded	Server is live and ready
failed	One of the stages failed — check logs

Build Pipelines

The Builds sub-section shows all Kaniko build jobs, including standalone builds you trigger manually.

See Build Pipelines for creating custom container images (e.g. custom Jupyter kernels with pre-installed packages).

Monitoring

Monitor active and completed workflows on the Dashboard. The activity feed shows recent generation job completions.