OpenHands v1.8.0 リリースOpenHands Releases v1.8.0

OpenHands, the open-source platform for building and running autonomous coding agents, has published version 1.8.0. The point release is modest in scope but centers on a change that gives users finer control over how their agents connect to external tools, alongside a fix that corrects how the system handles skills that a user has intentionally turned off. For teams that rely on agentic workflows, incremental releases like this matter because they shape the day-to-day reliability and configurability of the tools an agent can reach.

The headline feature in v1.8.0 is the ability to enable or disable an installed Model Context Protocol (MCP) server directly from its card in the interface. Previously, managing which MCP servers were active typically involved more manual configuration; contributing this card-level toggle, credited to the contributor hieptl in pull request #16171, lets users switch a server on or off without removing it entirely. That distinction is useful in practice: a developer can keep a server installed and configured while temporarily taking it out of the agent's active toolset, then re-enable it later without redoing setup.

MCP is the connective layer that makes this control meaningful. The Model Context Protocol is an open standard, introduced by Anthropic in late 2024, that defines a common way for AI applications to talk to external data sources and tools such as file systems, databases, issue trackers, and web services. Rather than writing bespoke integrations for each combination of model and tool, developers can expose capabilities through an MCP server that any compatible client can consume. OpenHands supports MCP servers as a way to extend what its agents can do, so the number and configuration of active servers directly affects the agent's capabilities and, potentially, its behavior. Being able to disable a server from its card gives users a straightforward lever to constrain or expand that surface.

The release also includes a bug fix, attributed to the contributor d3f4w2 in pull request #16168, that excludes disabled skills from the agent context. In OpenHands, skills are reusable units of functionality that inform how the agent approaches tasks. The issue, as described, was that skills a user had marked as disabled were still being included in the context passed to the agent. That behavior could lead the agent to consider or attempt actions the user had explicitly turned off, which undermines the purpose of disabling a skill in the first place. The fix ensures that disabled skills are omitted from the context, so the agent's working set more closely reflects the user's stated configuration. Reducing unnecessary items in the context can also help keep prompts focused, though the primary intent here appears to be correctness rather than optimization.

A smaller maintenance change in the release regenerates the favicon PNG and ICO assets from the source favicon.svg for a consistent design across formats. Changes of this kind are cosmetic and do not affect agent functionality, but they reflect the routine housekeeping that accompanies most point releases.

Taken together, the two substantive changes in v1.8.0 share a theme: giving users more precise and predictable control over what their agent has access to. Both the MCP server toggle and the disabled-skills fix are about aligning the agent's actual toolset with the user's configured intent, which is a recurring concern as agent platforms mature and users run them against real codebases and production systems.

The wider context is a fast-moving field of agentic development tools. OpenHands, which was previously known as OpenDevin, sits alongside a growing set of AI coding agents and assistants, and MCP has become an increasingly common integration point across the industry, with support appearing in a range of clients and tools since the standard was published. As adoption of MCP grows, the ability to manage individual servers cleanly is likely to become a more prominent part of the user experience, and the card-level enable/disable control in this release is a step in that direction.

For existing users, upgrading to v1.8.0 is a low-risk update focused on control and correctness rather than new capabilities or interface overhauls. Those who work with multiple MCP servers or who rely on selectively enabling and disabling skills stand to benefit most from the changes. As with any release, users should consult the project's changelog and repository for the complete list of merged pull requests and any upgrade notes relevant to their setup.