MCP is in private preview. To get access, ask your Photon contact.
1. Connect the agent
Point your agent’s MCP client at the Neutron server:https://ai.neutron.health. ⚠️ network.neutron.health/mcp isn’t live yet, so use ai.neutron.health for now. The agent authenticates like any other caller. A backend agent uses a machine token. An agent that runs in a prescriber’s MCP client, such as Claude or ChatGPT, signs in as that prescriber. See Connect a client.
2. Intake, search, draft
The agent calls three tools, in order:
Each tool’s inputs and outputs are in Prescribing tools. Photon makes the decisions here as well. When it isn’t sure which patient or drug is meant, the tool fails closed and tells the agent what to ask or send next. Agents should never invent ids, sigs or quantities.
If the agent learns something safety-relevant after drafting, such as a new allergy, it calls
rx_intake again and redrafts.
3. Hand the drafts to a prescriber
Pass the patient and the draft prescriptions to Prescribe. The prescriber sees each draft with its screening, signs, and sends.order="ord_…".
Sending from the agent
When the agent runs as the prescriber, it can send without Prescribe.rx_send takes the reviewed drafts and the prescriber’s explicit confirmation, and Photon checks on the server that the token belongs to a prescriber. A machine token can’t send. See rx_send.