IMTI

Every piece of the protocol has been on the table on its own.

The first two parts of this series ran entirely on a laptop.

The last primitive is the newest, and the only one the spec still marks experimental.

The authorization post controlled who may reach a server.

Over stdio the connection is trusted: the server is a local subprocess the client launched.

A server has two ways to tell a client something happened without being asked.

The primitives are done, the four the client calls and the three the server calls back.

Elicitation is the server asking a person for input in the middle of a task. elicitation/create runs server to client to user, the third and last of the inverted requests after sampling and roots.

Roots are the smallest of the server-to-client requests.

Generating an embedding is slow.

Every method so far has run one direction: the client asks, the server answers.

The series has deferred this method three times.

Prompts are the third thing a server can offer, after tools and resources.

Tools are actions the model takes.

The last post covered discovery: what the model reads to decide how to call a tool.

Discovery in MCP is one method, tools/list, and its response is the exact JSON the model reads to decide whether to call a tool and how to fill in its arguments.

Streamable HTTP is the MCP transport for servers reached over a network, behind auth, serving many clients at once.

The stdio transport is the simplest one MCP defines.

Wrap a real API in an MCP server and you hit a wall almost immediately.

Under 2025-11-25, a session opens with initialize, the server answers with what it can do, the client sends initialized, and only then do other methods become legal.

MCP is JSON-RPC 2.0 over a transport.

The common objection to MCP is that a REST API would do the same job.

You write a Go function, tag its arguments, and hand it to mcp.AddTool.

A pull request shouldn’t need four review rounds.

Building an MCP-powered data platform taught me that LLMs will ignore the instructions I spent months writing.

Moving analytics data from a Zalando-managed PostgreSQL cluster into OpenSearch for full-text search and dashboarding looks simple on paper.

GoReleaser automates building, packaging, and publishing Go binaries.

The principles article covered why verification matters.

My daughter called me an “unk” the other day.

Andrej Karpathy coined “vibe coding” in early 2025: “forget that the code even exists… I ‘Accept All’ always, I don’t read the diffs anymore.”

§Privacy Policy Effective Date: February 1, 2026 Industrial Media Technology, Inc.

§End User License Agreement Effective Date: February 1, 2026 This End User License Agreement (“Agreement”) governs your use of applications and services provided by Industrial Media Technology, Inc.

I’ve been experimenting with MCP for about six months now, building custom servers for data platform clients.

Data warehouses handle structured queries.

Data lives in PostgreSQL for transactions, MySQL for legacy systems, S3 for analytics, Kafka for streams.

kubectl port-forward works fine for one service.

kubefwd has been a stable and useful tool for my team and me for almost eight years.

kubefwd is a tool that bulk-forwards Kubernetes services to your local machine, making them accessible by their real hostnames.

I wrote portpxy to solve an annoying problem: NiFi lets you spin up HTTP listeners at runtime, but Kubernetes makes you update manifests every time you want to expose a new port.

This article covers advanced JOLT transformations including cardinality, modify operations, wildcards, and chained specifications for complex JSON restructuring.

This article introduces JOLT (JSON to JSON transformation Language) in Apache NiFi, covering the fundamental operations: shift, default, and remove.

This article covers securing Apache NiFi deployments with TLS encryption, authentication providers, and role-based access control, including LDAP, OIDC, and certificate-based authentication.

This article covers deploying Apache NiFi on Kubernetes for production workloads: a clustered deployment with ZooKeeper, persistent storage, and proper ingress handling.

This final article in the series covers high-performance computing techniques for linear algebra in Go: BLAS/LAPACK integration, parallel operations, memory optimization, and benchmarking.

This article implements neural network foundations in Go using gonum: a perceptron, forward propagation, and backpropagation from scratch.

This article implements Principal Component Analysis (PCA) from scratch in Go using gonum, covering both the covariance matrix and SVD approaches.

This article demonstrates building a regression library in Go from scratch using gonum: ordinary least squares, ridge regression, and cross-validation.

This article covers statistics and data analysis in Go using gonum/stat and gonum/mat: descriptive statistics, covariance matrices, and correlation analysis.

This article covers Singular Value Decomposition (SVD) and related matrix decompositions in Go.

This article covers eigenvalue problems in Go using the gonum library.