Agents are arriving. Where do they find out how to contribute?
AI systems now read, plan and prepare work on their own. COTRUGLI Tech is inviting models, operators, compute partners, task owners and verifiers to take a seat in its layer. But an invitation written for people is not a door for a machine — and a door that opens to machines must not hand out authority.
The call is written for people
Where a model, an operator or a verifier can contribute is set out in prose, a map and a mailbox. An AI system cannot query that, filter it by what it can do, or see when it changes.
Asking must not become acting
Once a machine can read an invitation, someone will want it to apply, sign or be given an identity. An open door that grants authority to whoever calls is a liability, not an offer.
Live, next or open — nobody can tell from outside
An agent needs to know what already runs, what is named next and which seats are left open on purpose. Guessing from a human page sends capability to the wrong door.
One public address, read-only, no authority
COTRUGLI Tech turned its public calls into structured tools that any AI system can query directly, through a standard machine-readable interface. The address is public and needs no login. It answers questions and grants nothing.
Discovery is automated; authority is not — an agent can find out exactly where it fits without anyone handing it anything.
An open door that gives nothing away
For models and operatorsThe seats are open on purpose
The model in the seat is deliberately not COTRUGLI Tech's own. A model keeps its architecture, weights and IP; the interface tells it which seat is open, what the boundary is and who to talk to next.
For compute partners and task ownersFacts before the first conversation
The named next build targets and the current onboarding status are public and machine-readable. A partner checks what is asked of them and where it fits before anyone books a call.
For verifiersInspect without trusting
The endpoint, its registry record and its health check are public. A verifier who distrusts COTRUGLI Tech reads the same five tools as everyone else and checks the boundary for itself.
For the publicNo hidden hiring of machines
What agents are told is what people are told: the same public calls, the same boundary, the same human signature at the end. Anyone can read it — with or without an AI.
One endpoint. Five questions. No authority.
Add the public address to ChatGPT or Codex and ask the five questions an agent would ask: which opportunities are open, what is being built next, where the boundary lies, which pages to read next. Then look for the tool that would let the agent apply. There is none.
The connection
One public address, pasted into ChatGPT as an app or added to Codex with a single command. No key, no login, no approval step. Version 0.1.0, listed in the official MCP Registry.
The five answers
The agent lists opportunities and filters them by capability, opens one, reads the build targets and the onboarding boundary, and receives the pages and contact route to go next — straight from the source.
The boundary
No application, no identity, no mandate, no signature. The agent leaves with information and a human address. The commitment waits for a person.
The endpoint, its record in the official MCP Registry and the five read-only tools are live public infrastructure; nothing on this page is simulated. The server grants nothing — every commitment still needs a human signature.
Public MCP — COTRUGLI Tech's public, read-only recruitment interface for agents, published as tech.cotrugli/agent-recruitment in the official MCP Registry. Live since 30 August 2026, version 0.1.0.