Typed access
MCP server
A typed surface over your operational substrate — customers, tickets, drafts, deals, whatever your real entities are. Per-user sk_studio_* keys, hash-stored, audit log on every call. See the five patterns →
§ 18Stack · Avira
Avira. The workspace pattern we built for our own program, now installed for one organisation at a time.
An MCP server tuned to your operational substrate. A skill library shaped to your work. The folder, citation, and governance conventions that turn a Claude Code session into a senior operator who has read everything you have. A small team. One engagement at a time.
[An Avira install: the 4–6 week engagement where we deploy the SoQ kit — an MCP server tuned to your substrate, a curated skill library, the folder/citation/governance conventions, and the studio-key auth surface — into your runtime.]
§ 01 / The problem we are solving
The model is the easy part. The scaffolding is what breaks.
The first week with an agent is fun. The second week is boilerplate. By the fourth week most teams discover that the model is the easy part — the hard part is everything around it: typed access to your real data, a skill spine someone curated, a place where context lives, a place where outputs go, a citation standard that keeps the work honest, and a refusal posture that does not embarrass you in front of a customer.
We know because we built one for our own program. An Avira workspace is the layer between us and our work — sales, parent communication, session synthesis, program design, fact-checking. The pattern is real. The artefacts compound with each cohort. We are productizing it now because every operator team we know is rebuilding the same scaffolding from scratch and getting halfway.
§ 02 / What is in an install
Five artefacts. One opinionated kit.
Curated workflows
Skill library
Five to nine workflows, named and shaped to your work. Same archetype as our /terra (fact-check) and /soqratic (review). Designed with you in week one.
Where context lives
Folder convention
A studio/ for inputs and reference. An artefacts/ for outputs. Dated, organised by entity. The simplest part of the kit and the part that breaks first when nobody owns it.
Honest output
Citation standard
Every non-trivial claim grounded with an inline citation. Inferences labelled as inferences. The thing that makes output legible to a regulated reader and impossible to skim past.
The part agencies skip
Governance bar
A CLAUDE.md with your operating principles, your refusal posture, and a pre-output quality gate. Without it, an agent ships work you would never have signed your name to. With it, the work is yours.
Operator handle
Admin surface
A small console for issuing and revoking studio keys, viewing usage, and rotating skills. Designed for one of your operators to own without paging an engineer.
§ 03 / How an install runs
Four to six weeks. The team in the room. Then you run it.
Week 0
Discovery
Two days on-site or remote. We read your substrate, sit with two of your operators, and list the work we would want a senior associate to do. We tell you if this is the wrong moment.
Weeks 1–3
Build
We scaffold the MCP, draft the skills, write the CLAUDE.md, ship the folder convention into your repo. You see a working version at the end of each week, not at the end of the engagement.
Week 4
Handoff
Two of your operators trained to extend the kit themselves. We leave a written maintenance plan and the playbook for adding the next skill. The kit belongs to you the day we leave.
After
Six-month note
An optional written check-in: what stuck, what slipped, what to add. If you want a longer relationship, the Fellowship lives under consulting.
§ 04 / Who this is for
Operators with a substrate worth typing.
- Teams that already have Claude Code or Cursor seats and a real substrate — a database, a Linear, a Drive, an internal app — worth typed access.
- Founders who want one operator doing the work of three, not three operators doing the work of one.
- Verticals where evidence-grounded output matters: education, legal, clinical, investment, regulatory. Anywhere "the inference looks plausible" is not enough.
- Organisations whose work has structure — students, cases, customers, drafts, reviews — not just "we have a lot of meetings."
§ 05 / What we don't do
Three honest non-promises.
We are not a connector aggregator. If you want one OAuth flow across Google Workspace, Microsoft 365, and Slack, that is a different product and a different company. We type your substrate — the database, the tools, the entities your work actually moves through. The third-party long tail is not the job.
We do not pick your model or your harness for you. We work where your team already works — Claude Code, Cursor, your internal harness, whatever you have. We bring the pattern, not the runtime.
We do not staff-aug.A small team. One engagement at a time. If we don't have capacity, we will tell you on the first call and recommend someone honest.
§ 06 / The shapes
Three sizes. One engagement model.
Avira is offered as enterprise engagements only. Pricing is part of the contract; per-domain curves are posted on request.
Install
Install
4–6 weeks · one domain
- One MCP server
- 5–9 curated skills
- Folder + citation + governance kit
- Two operators trained
Re-install
Re-install
2–3 weeks · second domain
- Reuses the harness
- New MCP surface for the new domain
- 3–6 new skills
- Cross-domain audit included
Sovereign
Sovereign
regulated & private
- Self-hosted MCP, your domain
- Compliance review tailored to your industry
- Custom skill development
- On-call workspace support
Bring your substrate. We will bring the pattern.
Write to foundersdesk@ with the actual situation and we will reply within three working days, from a person.