OzBridge is an open-source project that provides an unofficial bridge between the Warp Oz coding agent and various IDEs or MCP clients. It includes a Model Context Protocol (MCP) server and a VS Code extension, enabling MCP-compatible clients such as GitHub Copilot Chat, Claude Code, Cursor, and Codex CLI to run Warp Oz cloud and local agents. The project states that it is an independent project and is not affiliated with, endorsed by, or sponsored by Warp, Inc.; Warp and Oz are trademarks of Warp, Inc., used here only to describe interoperability. OzBridge uses only Warp's public interfaces (the oz CLI, the Model Context Protocol, and the WARP_OUTPUT_FORMAT environment variable), and does not modify, reverse engineer, or compete with Warp.
In terms of purpose, OzBridge is used to drive Warp Oz agents in any IDE or MCP client. It can run embedded in VS Code, making Oz appear as the @oz Chat Participant and Agent-Native Language Model Tools, which Copilot Agent mode can invoke autonomously; it can also expose the same Oz toolset over HTTP+SSE for Claude Code, Cursor, and Codex to call. If not using an editor, the @sena-labs/oz-mcp-server package can be run standalone, pointing any MCP client to it.
Key capabilities include: MCP bridge, exposing the Oz toolset over HTTP+SSE to any MCP client, running either embedded or fully standalone. @oz Chat Participant, interacting with the Warp Oz agent in the VS Code chat panel. Agent-Native Language Model Tools, allowing Copilot Agent mode to call Warp Oz directly without typing @oz. One-click model selection, via QuickPick (OzBridge: Select Model), the status bar indicator, or @oz /models <id> to select an Oz model; agents can switch via the MCP tools oz_list_models / oz_set_default_model. Sidebar and status bar, with an activity bar view containing Active Runs, History, Schedules, Environments, MCP Servers, and Secrets, plus active run and model indicators. Context variables and Warp handoff, expanding inline tokens such as #warp.env, #warp.profile, #warp.model, #oz.history, and #oz.run/<id> into /run or /cloud prompts, and supporting one-click handoff to a real Warp terminal. Per-workspace configuration, with an optional .warp/warp-bridge.yaml committed to the repository overriding ozBridge.* settings, with precedence YAML > VS Code settings > defaults, but mcpBearerToken and the platform-dependent ozPath are intentionally excluded. 9 slash commands covering the full agent workflow: /run, /cloud, /status, /history, /schedule, /models, /mcp, /config, /init. IDE context injection, automatically including the workspace path, active file, selection, and diagnostics in every prompt. Agent skill detection, mapping prompt keywords to a 7-agent pipeline (spec, design, implement, review, test, deploy, maintenance). Cloud run polling, using exponential backoff polling with real-time progress updates in the chat stream. Robust JSON parser, providing a 5-level fallback for mixed text/JSON CLI output. Configurable, with all settings exposed through ozBridge.* in the VS Code settings UI. Small runtime footprint, bundling only one workspace package (copilot-chat-toolkit) plus the vscode API, with no third-party network code, and a production bundle of about 160 KB.
In terms of usage, install from the registry: as of v0.9.0, every tagged release is synced to two registries. The VS Code Marketplace link is https://marketplace.visualstudio.com/items?itemName=sena-labs.ozbridge，Open The VSX link is https://open-vsx.org/extension/sena-labs/ozbridge。也可从 VSIX local installation: In the VS Code GUI, press Ctrl+Shift+P or Cmd+Shift+P, type "Extensions: Install from VSIX…" and select the ozbridge.vsix file; or use the CLI. After installation, press F5 in VS Code to launch the Extension Development Host. Installation verification: Open the Copilot Chat panel (Ctrl+Shift+I / Cmd+Shift+I), type @oz /config and submit it; the panel should display a table of the current configuration and the detected Oz CLI path; if the CLI is missing, an "Install Warp" action button will be displayed. Connecting an MCP client: First start the bridge, either by enabling the embedded server in VS Code ("ozBridge.mcpEnabled": true, then OzBridge MCP: Start), or by running the standalone package. Both listen by default on http://127.0.0.1:3847/sse。设置 ozBridge.mcpBearerToken can require an Authorization: Bearer <token> header. Client configuration examples include adding it to ~/.claude.json, ~/.cursor/mcp.json, ~/.codex/config.toml. VS Code requires no MCP configuration; after installing the extension, it can be used through the @oz chat participant and the Language Model Tools callable in Copilot Agent mode. Protocol details, endpoint reference, and curl quick reference are in docs/MCP.md.
Regarding applicable scenarios and limitations, it is applicable to macOS, Linux, and Windows, requires VS Code ≥ 1.101.0 (@oz participant requires the stable Chat Participant API; LM Tools also require vscode.lm.registerTool), Warp Terminal is installed and the oz CLI is in PATH, the Warp account is logged in via oz login, and the GitHub Copilot Chat extension is optional but is actually required when calling @oz or using Agent mode tools. Local runs start the oz CLI in the workspace, do not consume Warp credits, and have full file access; cloud runs execute on Warp infrastructure, consume credits, and run detached from the local machine. A free Warp account is sufficient for local agent runs; cloud runs may require a paid plan depending on usage. You can set ozBridge.defaultModel or specify the model inline via /run --model gpt-4o; use /models to view all available models. The default timeout is 300000 milliseconds (5 minutes), and can be increased to 600000 milliseconds for large-scale runs; cloud runs are controlled separately by cloudPollingTimeoutMs for longer timeouts. The extension activates only when the @oz participant is invoked in Copilot Chat or when Copilot Agent mode invokes LM Tools. For testing, there are 1400+ tests across 100+ files, with the framework being Vitest v4.0.18. The contribution process includes forking the repository, creating a feature branch, committing changes, pushing the branch, and opening a Pull Request; before submitting, ensure all tests pass; see also the Code of Conduct and Security Policy. The license is not provided in the materials.Before submitting, ensure all tests pass; see also the Code of Conduct and Security Policy. The license is not provided in the materials.
