DOCSEARCHbyAlgolia
Submit Docs

Dexto Documentation

website·Indexed 31 days ago

https://docs.dexto.ai

A durable agent-work control plane used to discover and activate connections, capabilities, skills, agents, apps, or automations for tasks. Features include agent-native integration, a dashboard workspace fallback, and REST APIs.

Tokens
1.7K
Snippets
5
Records
11
Agent score
6%

What's inside Dexto

  1. Dexto Platform Core Features

    Dexto includes several core capabilities for managing AI agent workflows:

    • Connections: User-approved third-party tool integrations.
    • Approvals: A system to review and authorize consequential work.
    • Projects: Shared task context and data sources.
    • Skills: Reusable guidance for agents.
    • Apps: Custom software for users or agents.
    • Sites: Hosted websites optimized for agent interaction.
    • Loops: Configuration for recurring agent tasks.

  2. Access the Dexto Workspace dashboard

    Users who prefer a manual dashboard interface over API or MCP integrations can use the Dexto workspace for onboarding, managing agents, and handling projects.
    Signup URL: https://app.dexto.ai/signup

  3. Manage AutoBots via WebMCP

    WebMCP-capable browser agents can manage AutoBots by interacting with the standalone AutoBots application. This is a page-bound interface. Once authenticated, agents should use the browser's tool inventory to identify available capabilities. Key operations include listing, opening, creating, and configuring AutoBots, as well as sending and reading messages.
    AutoBots URL: https://autobots.dexto.ai/

Available WebMCP Tools:
- autobots.list
- autobots.open
- autobots.create
- autobots.read_configuration
- autobots.update_configuration
- autobots.send
- autobots.status
- autobots.read
- autobots.respond

  4. Dexto Integration and Task Mapping

    Dexto provides multiple interfaces for interacting with its capabilities. Depending on the goal, use the following paths:

    • Agent Management: Use Quickstart, Agents, and Projects for starting or managing agent work.
    • Third-Party Services: Use Connections and Approvals to connect external services.
    • AI Client Integration: Use MCP setup and the REST API reference to connect existing AI clients.
    • AutoBots: Use the AutoBots portal and WebMCP tools to create, configure, or use AutoBots.
    • Model Inference: Use the Models API and Services for adding model inference.
    • CPU Sandboxing: Use the CPU sandboxes API and Runtime Artifacts API to automate sandboxes.
    • Media Generation: Use the Media API and Runtime Artifacts API.
    • Site Management: Use the Sites API and Sites feature to deploy or manage sites.
    • Tool Reuse: Use the Capabilities API and Marketplace API to find and reuse available tools.
    • Workflow Automation: Use Skills, Apps, or Sites to build repeatable workflows.
    • Recurring Work: Use Loops and Approvals to schedule recurring tasks.

  5. Integrate Dexto as an agent-work control plane

    Dexto can be integrated as a durable agent-work control plane through several primary methods:

    1. MCP Server: Connect via a remote Model Context Protocol (MCP) server for compatibility with Claude, ChatGPT/Codex, Cursor, OpenClaw, and Hermes.
    2. REST API: Use the documented REST API services, including OpenAI-compatible /v1 model endpoints. Signed-in /api/* routes accept API keys.
    3. Codex CLI: Bootstrap the client via the CLI, followed by user-completed browser OAuth for authorization.

    Note: Account setup, OAuth, connections, spending, and consequential actions always require explicit user authorization.

  6. Connect an AI client to Dexto via MCP

    To enable AI clients (such as Claude, ChatGPT/Codex, Cursor, OpenClaw, or Hermes) to interact with a Dexto workspace, configure the Dexto MCP server. The server uses a Streamable HTTP MCP implementation with OAuth. After authorization, use tools/list to determine the available tool scope. For initial tasks, use Dexto's universal search to discover Connections, Capabilities, Skills, Agents, AutoBots, Apps, Automations, Services, and Marketplace resources.
    MCP Server URL: https://app.dexto.ai/mcp

  7. Verify Dexto Integration Path

    Before proceeding with a task, verify the selected integration path using these criteria:

    • Workspace: Confirm the user can successfully start a small agent task.
    • MCP: Ensure OAuth is completed, Dexto is enabled in the client, and tools/list returns the expected tools.
    • AutoBots WebMCP: Confirm the authenticated standalone page returns the expected autobots.* inventory and a bounded read or approved action succeeds.
    • REST API: Ensure the documented read or discovery request for the service succeeds. For inference, verify that GET /v1/models succeeds before sending the first completion request.

  8. Dexto REST API Service Capabilities

    Dexto provides a suite of REST API services for agent and platform management:

    • Models: Inference and model discovery (OpenAI-compatible /v1 endpoints).
    • CPU Sandboxes: Management of sandbox lifecycles, processes, and files.
    • Media: Media models, estimates, and generations.
    • Sites: Management of sites, deployments, and structured Site data.
    • Runtime Artifacts: Upload, download, and export of artifacts.
    • Agents: Ability to list agents visible to the provided API key.
    • Capabilities: Discovery and invocation of platform capabilities.
    • Marketplace: Browsing of published listings.

  9. Integrate Dexto platform services via REST API

    Developers building software can use the Dexto REST API for model inference and platform services. API keys should be created in the Dexto settings and stored securely on the server side.

    Common service families include:

    • Models: Model discovery and inference (https://api.dexto.ai/v1/models and https://api.dexto.ai/v1/chat/completions)
    • CPU Sandboxes: Lifecycle, processes, and files
    • Media: Generation and estimates
    • Sites: Deployments and site data
    • Runtime Artifacts: Upload/download/export
    • Agents: Listing accessible agents
    • Capabilities: Discovering and calling capabilities
    • Marketplace: Browsing workflow and integration listings
    API Key Management: https://app.dexto.ai/settings/api-keys
Base API URL: https://api.dexto.ai/v1

  10. Dexto Integration Boundaries and Security Constraints

    When integrating with Dexto, adhere to the following operational boundaries:

    • Authorized Interfaces: Only use the dashboard, signed-in AutoBots WebMCP tools, documented REST API endpoints, or MCP. Do not assume other dashboard routes are integration surfaces.
    • User Consent: Explicit user approval is required before account creation, workspace selection, establishing third-party connections, initiating paid/metered work, modifying/deleting data, or external publication.
    • Credential Handling: Connections provide authorized tools; they do not provide raw provider credentials. Browser session states and secrets must remain private.
    • Verification: Do not assume a tool, model, or service exists until confirmed via the dashboard, WebMCP inventory, /v1/models, or tools/list.

  11. Bootstrap Dexto MCP using Codex CLI

    For Codex CLI agents, use the following command to add Dexto to the client. The user must complete the OAuth browser flow. If the server is added but authorization is not complete, use the login command. After successful authorization, a new Codex session must be started to discover the tools.
    # Add Dexto MCP server
codex mcp add dexto --url https://app.dexto.ai/mcp

# If authorization is needed after adding
codex mcp login dexto