{"slug": "agentdesktop", "title": "Agentdesktop", "summary": "Agentdesktop has launched an open-source platform that inventories AI developer tools, MCP servers, and skills across employee workstations, applies tool-native configuration and sandbox policy, and routes tools through an LLM gateway with user and device identity. The platform supports Claude Code, Codex, OpenCode, and VS Code, and can operate standalone or under a central controller for fleet management.", "body_md": "Agentdesktop discovers AI developer tools, inventories MCP servers and skills, applies tool-native configuration and sandbox policy, and connects each device to an LLM gateway with user and device identity.\n\nKeep developers in Claude Code, Codex, OpenCode, and VS Code while giving platform teams one place to understand and manage the fleet.\n\n[Website](https://agentdesktop.dev) ·\n[Documentation](https://agentdesktop.dev/docs/) ·\n[Announcement](https://agentdesktop.dev/blog/2026/09/introducing-agentdesktop/) ·\n[Releases](https://github.com/agentdesktop-dev/agentdesktop/releases)\n\nAI agents increasingly run on employee workstations, but the controls around them are fragmented across tool-specific settings, MCP connections, skills, provider credentials, and local configuration files.\n\nMDM remains the right layer for enrolling devices, deploying software, and enforcing OS posture. Agentdesktop adds the AI-tool-aware layer above it.\n\n| See what is running | Manage tools natively | Control model access | \n|---|---|---|\n| Discover supported tools and versions, then inventory MCP servers, skills, and models without collecting their secrets or contents. | Define configuration and sandbox intent once. Agentdesktop translates it into each supported tool's native format and reports whether it was applied. | Route tools through your LLM gateway with short-lived credentials carrying user, device, and allowed client context. Provider API keys remain at the gateway. | \n\nTry the same endpoint daemon used in a managed fleet without deploying a controller.\n\nDownload the current device binary from\n[GitHub Releases](https://github.com/agentdesktop-dev/agentdesktop/releases), or\nbuild it from source:\n\n```\ngit clone https://github.com/agentdesktop-dev/agentdesktop.git\ncd agentdesktop\ncorepack enable\nmake install\n```\n\nThe repository includes a working standalone configuration for Claude Code, OIDC, and Agentgateway. Preview every proposed file action without changing tool configuration:\n\n```\nagentdesktop daemon \\\n  --config examples/standalone/config.yaml \\\n  --user \\\n  --dry-run\n```\n\nRemove `--dry-run` to reconcile the configuration and leave the daemon\nrunning:\n\n```\nexport ANTHROPIC_API_KEY=sk-ant-...\ndocker compose -f examples/standalone/compose.yaml up -d\n\nagentdesktop daemon \\\n  --config examples/standalone/config.yaml \\\n  --user\n```\n\nIn another terminal, check the daemon and list discovered tools and models:\n\n```\nagentdesktop status\nagentdesktop discover\n```\n\nThe desktop and fleet interfaces also expose the MCP server and skill\ninventory. See the [standalone quickstart](https://agentdesktop.dev/docs/getting-started/standalone/)\nfor prerequisites, test credentials, and a walkthrough of the local services.\n\nAgentdesktop uses the same daemon and tool-native configuration model at every stage.\n\n| Standalone | Controller-managed | \n|---|---|\n| Read policy from local YAML, discover and configure tools on one workstation, and authenticate the user directly to a compatible LLM gateway. No controller or device identity is required. | Centrally inventory a fleet, distribute versioned configuration, enroll users and devices, report reconciliation status, and issue short-lived gateway credentials. | \n| [Run the standalone quickstart](https://agentdesktop.dev/docs/getting-started/standalone/) | [Run the managed quickstart](https://agentdesktop.dev/docs/getting-started/managed/) | \n\n- **AI tool discovery:** detect supported developer tools and their versions\nacross Linux, macOS, and Windows.\n- **Secret-minimizing inventory:** report configured MCP servers and skills\nwithout collecting MCP command arguments, environment variables, HTTP\nheaders, or skill bodies.\n- **Tool-native configuration:** safely merge managed values into the formats\nexpected by each tool while preserving unrelated user settings.\n- **Shared sandbox policy:** translate filesystem and network restrictions into\nthe native sandbox configuration supported by Claude Code and Codex.\n- **User and device identity:** bind a locally generated device key and\ncertificate to the user who enrolled the workstation through OIDC.\n- **Runtime credentials:** give supported tools short-lived credentials instead\nof distributing long-lived provider keys to workstations.\n- **Identity-aware gateway integration:** attach user, device, and allowed client\ncontext for gateway routing, policy, logging, and usage attribution.\n- **Opt-in telemetry:** collect selected session and tool-use events when an\norganization enables them.\n\n| Tool | Discovery | Managed configuration | MCP and skills inventory | Sandbox policy | \n|---|---|---|---|---|\n| Claude Code | Yes | Yes | MCP and skills | Yes | \n| Claude Desktop | Yes | Yes | MCP | — | \n| Codex | Yes | Yes | MCP and skills | Yes | \n| OpenCode | Yes | Yes | MCP | — | \n| VS Code | Yes | — | MCP and skills | — | \n\n**Don't see your tool?** We're actively expanding this list and would love\nyour help. [Open an integration request](https://github.com/agentdesktop-dev/agentdesktop/issues/new)\nto tell us what you use, or contribute discovery and configuration support\nfor another AI developer tool or harness.\n\nThe project targets Linux, macOS, and Windows. Support varies where a tool or operating system does not expose an equivalent native configuration surface.\n\nThe daemon runs on each workstation and reconciles developer-tool configuration. It can receive desired configuration from the controller or read the same YAML directly in standalone mode. Managed tools continue to run locally and can request credentials for an LLM gateway through the daemon.\n\nIn controller-managed mode, the daemon generates its private device key on the workstation and sends only a certificate signing request to the controller. After enrollment, protected controller operations require the device certificate and a valid token for the user who enrolled it.\n\nWhen a supported tool requests gateway access, the controller can issue a short-lived JWT containing the user, device ID, allowed client label, audience, issuer, and expiry. The client label is asserted by the local helper; it is useful for policy and attribution, but it is not cryptographic proof of the calling executable.\n\nRead the [announcement](https://agentdesktop.dev/blog/2026/09/introducing-agentdesktop/)\nfor a deeper walkthrough of standalone mode, enrollment, and short-lived tool\ncredentials.\n\nAgentdesktop is fully open source under the [Apache License 2.0](/agentdesktop-dev/agentdesktop/blob/main/LICENSE).\n\n- Read the [documentation](https://agentdesktop.dev/docs/) .\n- Browse or report [issues](https://github.com/agentdesktop-dev/agentdesktop/issues) .\n- Help us [add support for another AI developer tool](https://github.com/agentdesktop-dev/agentdesktop/issues/new) .\n- Review the [Code of Conduct](/agentdesktop-dev/agentdesktop/blob/main/CODE_OF_CONDUCT.md) before contributing.\n- See the [production guide](https://agentdesktop.dev/docs/operations/production/) for Kubernetes, certificates, MDM, and endpoint enrollment.", "url": "https://wpnews.pro/news/agentdesktop", "canonical_source": "https://github.com/agentdesktop-dev/agentdesktop", "published_at": "2026-09-08 01:15:25+00:00", "updated_at": "2026-09-08 01:31:30.911278+00:00", "lang": "en", "topics": ["developer-tools", "ai-tools", "ai-infrastructure", "ai-agents"], "entities": ["Agentdesktop", "Claude Code", "Codex", "OpenCode", "VS Code", "GitHub", "Agentgateway"], "alternates": {"html": "https://wpnews.pro/news/agentdesktop", "markdown": "https://wpnews.pro/news/agentdesktop.md", "text": "https://wpnews.pro/news/agentdesktop.txt", "jsonld": "https://wpnews.pro/news/agentdesktop.jsonld"}}