DeepSeek Harness v0.1.1 Released with Multimodal Support and MCP/ACP Integration
DeepSeek-Harness v0.1.1-rc.1 has been released, adding support for the new DeepSeek-V4-Flash-Vision-Exp multimodal model. This update also introduces native image request configurations and persistent image attachments for MCP/ACP. This release enhances the capabilities of DeepSeek models by enabling advanced visual understanding and seamless integration with external tools and agents. It allows developers to build more interactive, multimodal AI workflows using standardized open protocols like MCP and ACP. Commands such as /goal and /plan can now accept both text and image inputs, while the @ menu allows referencing files and sessions. Additionally, PTC Mode now supports forwarding nested images, and MCP/ACP supports persistent image attachments.
## BACKGROUND
The Model Context Protocol (MCP) is an open standard introduced by Anthropic to standardize how AI models connect with external data sources and tools. Similarly, the Agent Connect Protocol (ACP) is designed to enable interoperability and communication between different AI agents across various frameworks.