Standard
  • Docs
  • Examples
  • Pricing
  • About

Get started

  • Overview
  • Conversation analytics
  • How Standard works
  • Run your first task
  • Connect
  • Join a Space

Build

  • Authoring · Sales calculation
  • Examples
  • Software release verification
  • Operation profiles
  • Integration reference

Concepts

  • Sources and customer vaults
  • Inputs and permissions
  • Accounts and billing
  • Results and lifecycle
  • Verification checks
  • Follow-on tasks
  • Cleanup

Security

  • Security model
  • Threat model and AI risk mapping
Documentation menu

Get started

  • Overview
  • Conversation analytics
  • How Standard works
  • Run your first task
  • Connect
  • Join a Space

Build

  • Authoring · Sales calculation
  • Examples
  • Software release verification
  • Operation profiles
  • Integration reference

Concepts

  • Sources and customer vaults
  • Inputs and permissions
  • Accounts and billing
  • Results and lifecycle
  • Verification checks
  • Follow-on tasks
  • Cleanup

Security

  • Security model
  • Threat model and AI risk mapping
Standard

Managed confidential execution for approved functions.

© 2026 Standard Runtime, Inc. All rights reserved.

  • Security
  • Docs
  • Examples
  • Pricing
  • FAQ
  • About
  • Contact
  • Terms
  • Privacy
  • Cookies
  • Docs
  • Examples
  • Pricing
  • About
  • Get started

    • Overview
    • Conversation analytics
    • How Standard works
    • Run your first task
    • Connect
    • Join a Space

    Build

    • Authoring · Sales calculation
    • Examples
    • Software release verification
    • Operation profiles
    • Integration reference

    Concepts

    • Sources and customer vaults
    • Inputs and permissions
    • Accounts and billing
    • Results and lifecycle
    • Verification checks
    • Follow-on tasks
    • Cleanup

    Security

    • Security model
    • Threat model and AI risk mapping

    Get started

    • Overview
    • Conversation analytics
    • How Standard works
    • Run your first task
    • Connect
    • Join a Space

    Build

    • Authoring · Sales calculation
    • Examples
    • Software release verification
    • Operation profiles
    • Integration reference

    Concepts

    • Sources and customer vaults
    • Inputs and permissions
    • Accounts and billing
    • Results and lifecycle
    • Verification checks
    • Follow-on tasks
    • Cleanup

    Security

    • Security model
    • Threat model and AI risk mapping

    Connect

    Sign in from the CLI, authorize the SDK from your code, or connect an agent through MCP; every entry point reaches the same account and operations.

    Connecting establishes account access before Run your first task. Choose the entry point that matches where the request comes from. The interfaces share authority requirements, but their installation, custody and execution capabilities differ. Installed clients verify delivered result bytes; hosted interfaces return delivery metadata, and the authorized recipient integration verifies the result on its delivery path. Connecting grants no authority: each required owner still approves what runs and what leaves.

    Before you connect

    Complete assisted setup through Request access. Agree:

    1. The execution route available to your account and which participant hosts and recipients it needs. Connecting alone does not establish execution readiness.
    2. How to obtain and verify the client bundle, install it on its supported Linux x86_64 host, and configure the CLI endpoints and account binding.
    3. Enrollment of each required owner host, its key custody, Space participation and approval workflow. Local MCP requires that enrolled CLI host.
    4. The recipient’s registration, delivery access and independent verification references.
    5. For local function authoring, Docker and the exact preparation image pinned by the client bundle. The CLI uses a locally present image; Standard does not distribute it through these docs. Agree its supply during setup.

    The Node.js SDK invokes the installed CLI; importing the SDK does not install or enroll that CLI. The SDK and local MCP server are supplied with the client bundle. These pages do not provide a public package-install command or a self-service enrollment recipe.

    Connection and execution availability

    Hosted MCP can authenticate, discover tools and originate configured hosted requests. That does not make it equivalent to an installed owner’s execution path. After admission approval, if Control has not reported progress, hosted status returns hosted_continuation_unavailable: the continuation transport is not available on that path. Further polling does not start execution. Confirm the supported route during assisted setup.

    Connected-source CLI requests currently require hosted custody; an installed owner’s local-file setup cannot simply be changed to --source. See Connected-source route.

    CLI

    Use the CLI to deploy a function with its profile, run operations, receive verified results and inspect evidence from a terminal or a script.

    bash
    standard login

    Sign-in opens your browser. Sign in to your Standard account, review the CLI that is requesting access and approve it. The CLI keeps its credentials in your operating system's credential store, and standard status reports local credential storage, expiry and refresh uncertainty. It does not identify the signed-in account or establish that the provider still accepts the credential.

    PreviousRun your first taskNextJoin a Space

    On this page

    • Before you connect
    • Connection and execution availability
    • CLI
    • SDK
    • MCP
    • Hosted server
    • Local MCP server
    • Next step

    SDK

    Use the SDK when an application, CI/CD job or backend service submits operations from its own code. It authorizes with the same account as the CLI. As an installed client it verifies delivered result bytes in your environment before a result reaches your code; hosted delivery returns metadata that the authorized recipient integration verifies on its own path.

    ts
    import { Standard } from '@standardruntime/sdk';
    
    const standard = new Standard();

    standard.run() and operation.receive() do what standard run and standard receive do, with the same authorization and the same checks. Publication differs: standard.deploy() publishes a profile that runs Wasm only when it is given the build report and fixtures from standard.build() or standard.test(), and is refused with report_required otherwise. Use the CLI's standard deploy with --report and --tests to publish.

    MCP

    Standard has two MCP servers with different roles.

    • Hosted server: for remote agents such as ChatGPT, with nothing to install. The agent originates configured requests within its authority, subject to the hosted continuation limit. Each required owner approves the operation in the Console with a passkey. Hosted MCP returns delivery metadata; the authorized receiving integration retrieves and verifies the permitted result.
    • Local server, standard-mcp: for agents on the same host as the CLI, such as Claude Code or Cursor. standard_build and standard_test prepare a function locally with fixture data and use no account or key. standard_run, standard_status and standard_receive need a host enrolled with the CLI and use that host’s own keys. Publication is not an MCP tool: the person runs standard deploy with the build report. See Local MCP server.

    Hosted server

    • Endpoint: https://standardruntime.com/api/mcp
    • Transport: Streamable HTTP
    • Authorization: OAuth, discovered automatically from the endpoint

    Add the server to your client

    In clients that add remote servers through settings, such as a custom connector or integration, enter the endpoint URL. Leave any authentication fields empty: the client discovers OAuth from the endpoint.

    In clients configured with a JSON file, add an HTTP server entry:

    json
    {
      "mcpServers": {
        "standard": {
          "type": "http",
          "url": "https://standardruntime.com/api/mcp"
        }
      }
    }

    The exact file and key names depend on your client. Use its documentation for adding a remote Streamable HTTP server.

    Authorize your account

    The first time the client connects, it opens a browser window:

    1. Sign in to your Standard account. Accounts require multi-factor authentication.
    2. Review the client that is requesting access, then approve it.
    3. Return to your client. It now lists Standard's tools.

    Authorizing a client lets it act within your account's permissions. It does not grant access to another owner's inputs, and it does not authorize a payment.

    Check the connection

    Ask your agent what Standard can do for this account. It calls Standard's getting-started tool, which returns your account, your Spaces and the next steps available to you. If the client cannot connect, confirm that it supports remote Streamable HTTP servers with OAuth.

    Never paste private inputs, keys or another owner's evidence into the conversation or into tool arguments. Your agent passes the references accepted by its transport. An installed participant host prepares local encrypted contributions; confidential ingress retrieves connected-source material on a configured execution path.

    Local MCP server

    standard-mcp ships with the SDK package @standardruntime/sdk. It is a stdio server, so the agent starts it on the same host as the CLI. Each tool runs the matching CLI command and returns the CLI's result. Local preparation (standard_build, standard_test) uses no account or key; for standard_run, standard_status and standard_receive, sign in and enroll that host with the CLI first, and those tools use that host's keys. Connecting grants no authority. No tool approves: an operation without the required owner approvals waits or refuses, and owners approve in the Console.

    Next step

    With a connection in place, deploy the function and run an operation. Set up protected storage when the operation needs confidential inputs.