A/02Accelerators
MCP Server Kit
Give agents narrow, typed tools — never a shell.
The problem it solves
The fastest way to connect an agent to a system is one generic tool: run a command, query anything. It demos well and fails every security review, because the agent can then reach whatever the model decides. The real work is turning a runbook into narrow tools, and that work repeats from project to project.
A starting point for Model Context Protocol servers with a small, typed, read-first tool surface, a log of every call, and an approval step for anything that changes state.
What’s inside
- ✦A server scaffold where each tool is a typed function with bounded inputs
- ✦Read-only tools by default; changes go through an explicit human approval
- ✦A record of every call: who asked, which tool, what came back
- ✦Path and secret checks at the tool boundary for anything that touches files
- ✦Rate limits, so a retrying agent cannot flood the systems behind it
How a project uses it
- 01Walk through the runbook your engineers already follow
- 02Turn each step into a tool, and write down what is deliberately left out
- 03Run it against real cases outside production
- 04Hand over the tool list as a reviewed, documented scope
What it is not
- A general-purpose agent
- Unattended write access to production
- A way around your change-management process
Works well with
- A/04
LLM Evaluation Harness
“Is it better?” — with an answer that is a number.
Regression testing for prompts, models and retrieval: a versioned set of cases drawn from your own history, scored on every change.
- Evaluation
- CI
- Regression testing
- Model comparison
- A/01
Governed RAG Foundation
Answers from your documents — only from the ones the person asking may read.
Retrieval-augmented generation over company data, with access control carried from the source system and a citation on every answer.
- RAG
- Vector search
- Access control
- Citations
Start a project
Want this under your project?
Tell us what you are building and what you already have. We will tell you whether this accelerator fits, what it would need adapting, and where it would not help.
No form and no automated follow-up sequence. An email reaches an engineer, and the first reply is either a real question or an honest “not us.” We work in English, remotely, across European time zones.
We reply to every message that isn’t automated.