Ollama, a tool for running large language models (LLMs) locally, has announced that Claude Code, Anthropic's coding tool, can now use subagents and web search when running on Ollama[1]. Neither feature requires a separate MCP server or API key; you simply specify a model on Ollama and launch Claude Code[1]. The update lets you process several tasks in parallel and have the model search for up-to-date information on its own, helping make longer coding sessions more efficient.

Running Tasks in Parallel with Subagents

Subagents run tasks such as file search, code exploration, and research in parallel, each in its own separate context[1]. Because these side tasks proceed apart from the main work, the main context does not fill up with extra noise, and longer coding sessions stay productive[1].

Some models call subagents automatically when needed, and you can also start them explicitly by telling the model to "use" or "spawn" subagents[1]. Ollama lists minimax-m2.5, glm-5, and kimi-k2.5 as models that tend to trigger subagents naturally[1].

Example prompts include having subagents explore the authentication flow, payment integration, and notification system separately, or auditing security issues, performance bottlenecks, and accessibility in parallel[1].

Web Search Built In

Web search is now built into the Anthropic compatibility layer that Ollama provides (the mechanism that communicates in the same format as Anthropic's API)[1]. When a model needs current information, Ollama runs the search and returns the results directly, so users do not have to make any additional configuration[1].

Combined with subagents, you can research several topics in parallel and have the results pulled together into something you can act on[1]. For example, a single instruction can look up a piece of software's release notes, check your own code for outdated patterns, and create migration tasks[1]. Fetching this kind of external information sometimes required setting up an MCP (Model Context Protocol) server, but this update removes that step[1].

How to Use It and Recommended Models

To get started, specify a model and launch Claude Code[1].

ollama launch claude --model minimax-m2.5:cloud

The feature works with any model on Ollama's cloud[1]. The cloud models that Ollama recommends for coding are minimax-m2.5:cloud, glm-5:cloud, and kimi-k2.5:cloud[1].

The basic mechanism for running Claude Code on Ollama comes from the existing "ollama launch" command, which starts various coding tools without environment variables or configuration files[2]. The new subagents and web search are features added on top of that foundation.

Summary

Ollama has made subagents and web search available in Claude Code. Subagents process tasks in parallel within separate contexts to keep the main context organized, while web search is built into the Anthropic compatibility layer and retrieves current information with no extra setup. Neither requires a separate MCP server or API key; you just specify a model with "ollama launch claude." For anyone who wants to use Claude Code more seriously on their own hardware or with open models, this is an update that widens the options.

Source: https://ollama.com/blog/web-search-subagents-claude-code

Source: https://ollama.com/blog/launch