Class DockerExecClient

src/benchmarks/swe_bench_verified/docker_exec_client.py:23–367

MCP-compatible client that routes tool calls through `docker exec`. Subclasses MCPClient with connect=False to skip HTTP connection, then overrides invoke, list_tools, close, and get_session_id to use docker exec against an official SWE-bench container.

class DockerExecClient(MCPClient):
    """MCP-compatible client that routes tool calls through `docker exec`.

    Subclasses MCPClient with connect=False to skip HTTP connection,
    then overrides invoke, list_tools, close, and get_session_id to
    use docker exec against an official SWE-bench container.
    """

    def __init__(
        self,
        image: str,
        container_id: Optional[str] = None,
    ):
        """Initialize DockerExecClient.

        Args:
            image: SWE-bench Docker image name (e.g. swebench/sweb.eval.x86_64.django_1776_django-10880:latest)
            container_id: Optional existing container ID for resume.
        """
        # Initialize parent without connecting to MCP server
        super().__init__(connect=False)

        self.image = image

        if container_id:
            self._container_id = container_id
        else:
            # Start a new container from the image
            # -w /testbed sets the default working directory (matches mini-swe-agent)
            result = subprocess.run(
                [
                    "docker",
                    "run",
                    "-d",
                    "--rm",
                    "-w",
                    "/testbed",
                    image,
                    "sleep",
                    "7200",
                ],
                capture_output=True,
                text=True,
                timeout=120,
            )
            if result.returncode != 0:
                raise RuntimeError(
                    f"Failed to start container from {image}: {result.stderr}"
                )
            self._container_id = result.stdout.strip()

        self._session_id = self._container_id
        print(f"[DockerExecClient] Container started: {self._container_id[:12]}")

        # Run eval-compatible setup (matches the SWE-bench eval harness script)
        if not container_id:
            self._setup_container()