According to a recent LinkedIn post from Sourcegraph, the company is highlighting a new Code Finder capability embedded in its Sourcegraph MCP Server, now available in beta. The post describes Code Finder as an agentic search tool that autonomously scans code repositories, returning specific file paths and line ranges with brief descriptions to support coding agents.
The LinkedIn post suggests that Sourcegraph benchmarked Code Finder against alternative approaches, including a coding agent using local search and one calling existing MCP tools. According to the post, Code Finder completed file-finding tasks more than twice as fast as local search and at up to 40% lower cost than agents using MCP tools, with comparable result quality.
As outlined in the post, the performance gains are attributed to a division-of-labor model in which Code Finder is optimized solely for repository search, while general-purpose coding agents focus on reasoning. This specialization is presented as particularly beneficial for large codebases, where search overhead can consume a significant share of an agent’s token and cost budget.
The post also notes that Code Finder is free during its beta period and already powers file discovery within Sourcegraph’s Deep Search product. By reducing the token budget required to locate files, the feature is positioned as enabling Deep Search and external agents to devote more resources to higher-value analysis and reasoning tasks.
For investors, the post indicates ongoing product innovation in Sourcegraph’s developer tooling and AI-agent integration stack, which may enhance the platform’s value proposition for enterprises managing large codebases. If the reported speed and cost efficiencies translate into broader adoption, Sourcegraph could strengthen its competitive position in code intelligence and AI-assisted development, potentially supporting subscription growth and deeper customer integration.