Sign in as yourself

Sign in with your own NewInstance account and let an AI agent work in your organizations as you. There is no API key to create, copy or store: the agent connects, you sign in once in your browser, and it can then do what you can do.

Let your agent set itself uplink

Give your agent this one line. It reads the setup page, adds the server for the client it runs in, and walks you through signing in:

Read https://service.newinstance.cloud/mcp/setup and follow it to set up and connect the NewInstance MCP server for me.

The setup page is plain text at https://service.newinstance.cloud/mcp/setup. It covers Claude Code, Cursor, Claude Desktop and claude.ai, VS Code, Windsurf, Codex CLI and Gemini CLI, and has troubleshooting for sign-in problems. The manual steps below do the same thing by hand.

One endpoint ({{serviceUrl}}/mcp):

https://service.newinstance.cloud/mcp

The agent connects with no key. The server answers 401 with a WWW-Authenticate header, and any MCP client that supports sign-in follows it through the standard MCP OAuth flow for you.

Add it to your agentlink

Claude Code

claude mcp add --transport http newinstance https://service.newinstance.cloud/mcp

Then run /mcp in Claude Code, choose newinstance and select Authenticate to start the sign-in.

Cursor: add an entry to ~/.cursor/mcp.json (every project) or to .cursor/mcp.json in one project:

{ "mcpServers": { "newinstance": {
    "url": "https://service.newinstance.cloud/mcp" } } }

claude.ai: open Settings, then Connectors, choose Add custom connector and enter the same URL.

What happens the first time you connectlink

1. Your browser opens.
2. You sign in with your NewInstance account.
3. You approve the app on the consent page. The page says the app can do anything you can do in every organization you belong to, and that changes happen immediately. You approve once per app.

What the agent can dolink

What you can do in each organization you belong to, apart from the things listed under What the agent cannot do. It works through seven tools:

Every tool except list_organizations takes an orgId. Get it from list_organizations.

Files and PDFslink

The agent can upload a file in three steps:

1. It asks for an upload address with requestPresignedUpload.
2. It sends the file's bytes to that address with an HTTP PUT.
3. It calls confirmFileUpload. The file is then ready to use, for example as your Docs Portal logo or favicon.

It gets a file's download address with getFileUrl and deletes a file with deleteFile. A file uploaded into an EHR folder (a path starting with ehr/) is always stored privately, and a file that was not stored privately cannot be attached to a patient record.

An invoice from myInvoicePdf comes back as a PDF file attached to the result, not as text. EHR PDFs (a document, a treatment plan and a client face sheet) come back as a short-lived download address.

Patient recordslink

In the EHR the agent works with clients, encounters, documents, treatment plans, scheduling, billing and claims. Your own EHR permissions apply, and so do the consent rules (42 CFR Part 2), the break-glass rules and access logging. Every record the agent opens is logged as you opening it.

What the agent reads goes to the AI provider behind it. Before you connect an agent to an organization that holds patient records, your organization must be satisfied that this provider may receive that health information, for example under a business associate agreement. This is your organization's compliance decision.

Let your agents use keys and secretslink

In the dashboard, open Account settings, then the Connected agents tab, and turn on key access: Let my agents create and see keys and secrets. It is off by default. Only you can change it. An agent cannot read or change it.

• It applies to all of your agents.
• It is checked on every call, so turning it off takes effect on each agent's next call.
• When it is on, your agents can do everything listed under Blocked unless you turn on key access below. They see secret-style values in full instead of [hidden], and they can set secret values such as your support portal's sign-in secrets, webhook secrets and PINs.
• A setup plan (run_setup_plan) that includes a change that creates, rotates or reveals a key, token or secret is refused as a whole, and nothing in it runs. The agent runs each of those changes on its own with run_mutation.
• Every secret an agent reads or creates is recorded in your audit log, reads included.

Keys and tokens outlive the agent. Keys and tokens an agent creates keep working after you disconnect the agent or turn the setting off. Revoke them where they were created.

What the agent cannot dolink

This is the complete list of what is blocked. Everything else is limited only by your own permissions.

Always blockedlink

• Acting as your customers. Sign in to your support portal as a customer, with a password, an SSO or third-party sign-in token, or a sign-in link, or create tickets, comments and ticket attachments the way a customer does. Use the customer side of the chat widget. Use the public side of your docs portal: visitor views and the API playground.
• Your account and other agents. Sign out, open your account management link (it hands out a sign-in link), accept an organization invitation, register or remove push notification devices, approve, deny, list or disconnect agent connections, and read or change the key access setting. It also cannot use the dashboard assistant or its saved conversations.
• Billing. Report billing usage.

Blocked unless you turn on key accesslink

• Keys and tokens. Create or regenerate API keys. Create or rotate BugWatch project keys and mobile secrets, finish a BugWatch project draft (that hands out a key), or see BugWatch project keys and mobile credentials. Rotate your support portal's customer token signing secret. Create chat customer tokens or docs preview tokens.
• EHR. Create an EHR webhook subscription (it returns the signing secret), rotate an EHR webhook secret, set your EHR signature PIN, or save SMS provider credentials for appointment reminders.
• Secret Manager. Read the master key, its check value or the inheritance link key. Set up a master key for an app, or link inherited values. Secret Manager values are stored as ciphertext and the agent never receives the master key, so it cannot read them. It can still write values as ciphertext.
• Secret values it reads. Anything stored under a secret-style name (secrets, passwords, tokens, API keys, private and signing keys, salts), and any header-style value (Authorization, Cookie and API key headers), shows as [hidden] in everything the agent reads. If the agent sends [hidden] back as a value, the call is refused.
• Secret values it sets. A change that supplies one of these is refused: your support portal's sign-in secrets and salts (the general secret, the SSO secret key and salt, and the external sign-in secret key and salt), the support portal's chat widget key, webhook secrets such as the Support Tickets webhook secret, PINs such as your EHR signature PIN, and header-style values (Authorization, Cookie and API key headers). Set them yourself in the dashboard. Other values are not checked this way, so the agent can still set, for example, your email provider credentials.
• Sign-in tests and previews. Test your portal's third-party sign-in with a sample token. Open draft previews of your docs portal.

Changes run immediatelylink

A change runs the moment the agent makes it. Your agent app's own confirmation prompt for tool calls is the check, so keep it switched on for run_mutation and run_setup_plan.

See and disconnect your agentslink

Open Account settings, then the Connected agents tab. Choose Disconnect next to an agent and confirm. It stops working immediately.

An agent acting as you can change settings that keep working after you disconnect it. These are who your support portal trusts for customer sign-in (the sign-in endpoint, its public key, and the external and SSO sign-in URLs), webhook URLs, your email provider, and team invitations. With key access on, the keys and tokens it creates keep working too. These changes are visible in the dashboard and recorded in the audit log, so after you disconnect an agent you no longer trust, review those settings and your team members, and revoke any keys and tokens it created where they were created.

Servers and automationlink

For server-to-server use, keep using an API key as described on the AI Agent Access (MCP) page. That method is unchanged.