IRMA for developers
This page is for developers, IT administrators and anyone who wants an AI agent (such as Claude Code) to work with IRMA. It covers what works today, what is coming and where to start. If you work in the app yourself, read IRMA: getting started.
What you can do
Section titled “What you can do”- Report an incident. Your monitoring, SIEM or service desk files an incident in IRMA’s Incidenten (incidents) register automatically. A repeated report of the same problem does not become a second incident: IRMA counts it on the existing one. This works today, with a service token.
- Read what is in IRMA. Your working context, risks with their gross and net score, the change history of a risk, what changed in the risk register since the import, Monte Carlo simulations and how two compare, the coverage of a framework, the audit checks with their completeness score, your own registers and the incident register.
- Start a simulation. A new Monte Carlo run with a fixed seed, so the result can be reproduced.
- Propose drafts (later). An agent proposes changes as a draft; a person accepts them. See Drafts, not direct changes.
Four doors
Section titled “Four doors”IRMA opens up in this order: first the chat, then the API with the CLI, then an MCP server. All doors use the same task logic inside IRMA, so the same answers and the same permissions.
| Door | For whom | Status |
|---|---|---|
| The IRMA assistant in chat | Everyone with IRMA, in the Prudai web app. | Available. See The personas. |
| The /v1 API (task layer) | Developers and agents writing their own code. | Available, with the token of a signed-in user; scopes for service tokens are coming. |
The irma CLI | Scripts, cron and coding agents. | Available; a public install package is still to come. |
| An MCP server | AI clients that speak MCP (Claude, Cursor and similar). | Later. |
Separate from these is the machine plane /agent/v1: the door for systems that report incidents. It exists today and stays.
Where to start
Section titled “Where to start”- Quickstart: create a service token, check it and report your first incident.
- Concepts: organization and read scope, the fixed answer shape (the envelope), permissions and scopes, and what is recorded about an agent.
- Recipes: connect a SIEM or service desk, compare two simulations, changes since the import, an agent that prepares an audit.
- API reference (Prudai staff only, sign-in required): every
/v1endpoint, with examples. - CLI reference: every command of the
irmaCLI. - Agents: the Claude Code skill,
llms.txt, Markdown versions of these pages, and the MCP server that comes later.
Agreements that do not change
Section titled “Agreements that do not change”- Your organization, your permissions. A key or agent never sees more than its role and read scope in IRMA allow. A refused action is the design, not a bug.
- Honest answers. Every answer of the /v1 API says how complete it is, what is not in it and where the numbers come from. See The envelope.
- Additions only.
/v1only changes by addition: new fields and endpoints, never a field that disappears or changes meaning. Every change goes into a changelog. - Agents write drafts, with one exception: reporting incidents.
API reference
Section titled “API reference”The IRMA API reference (Prudai staff only, sign-in required) describes every /v1 endpoint with its parameters, fields, error codes and examples in curl and the irma CLI. The rest of the IRMA API is internal and is not published. See also API & integrations.