What Is MCP for NDIS Support Coordination?

Model Context Protocol, usually shortened to MCP, is an open protocol that gives AI applications a consistent way to discover and use tools. In support coordination software, that can turn an AI assistant from a chat window into an agent that can carry out a defined job.

The useful question is not simply, "Does this software have MCP?" The useful question is, "Which jobs can the agent do, whose access does it inherit, and what evidence is left after it acts?"

What MCP changes for a support coordination team

A normal AI chat can draft text from information placed in the conversation. An MCP connection can expose named software tools. A compatible agent may be able to find a participant, create a case note, add a follow-up task, review unbilled work or inspect an invoice.

That can reduce repeated navigation and rekeying, but it also creates a stronger security requirement. The agent is no longer only suggesting words. It can request actions in the system of record.

A practical CordoCare MCP example

CordoCare provides an organisation-scoped endpoint at https://cordocare.com/api/mcp. An owner or admin enables access for their organisation. An employee then creates a named personal key for a compatible agent connection.

The employee key does not become a master organisation credential. CordoCare recalculates the employee's live membership, role, team and participant access on every request. A coordinator agent stays within coordinator permissions. A finance agent only reaches finance tools when that person's role allows it.

Examples of exposed work

list or open authorised participant records
create or update participant information within the user's scope
create case notes and follow-up tasks
review unbilled notes, billing runs and invoices when permitted
perform limited team administration for authorised owners or admins
read recent audit history and roll back supported actions

Five controls to look for in an NDIS MCP

1. Organisation ownership

Each organisation should decide whether agent access is enabled. One customer must never be able to discover or act on another customer's records.

2. A real employee identity

Personal agent access should map to an active employee rather than a shared team password. When that employee is deactivated or changes role, the agent's effective access should change immediately.

3. Minimum useful scopes

Read, participant write, note write, task write and billing permissions should be separable. The agent should receive the least access needed for its job.

4. Repeat-safe writes and audit evidence

Agents can retry requests. Idempotency controls stop a retry from creating the same record twice. Audit history should identify the person, agent key and action.

5. Fast revocation

Employees should be able to revoke a connection, and the organisation should have a kill switch. Expiring keys reduce the life of forgotten access.

What MCP does not decide

MCP standardises a connection pattern. It does not decide whether a tool is professionally appropriate, whether participant consent has been handled, or whether AI output is accurate. Those decisions remain with the organisation and its people.

The official MCP authorisation specification emphasises restricted tokens and resource-specific access. In practice, an NDIS organisation should also apply privacy, record-keeping and human-review rules to the workflow around the connection.

A good first agent workflow

Start with a narrow, reversible job. For example: find an assigned participant, summarise the recent authorised context, prepare a draft follow-up task and ask the employee to approve it. Review the audit trail, then expand only when the boundary works as expected.

CordoCare MCP is included in the Practice plan. See the CordoCare AI agent access model or review the complete agent feature list.