As models improve as agents, we've found success by providing fewer details up front, making it easier for the agent to pull relevant context on its own.

Cursor

Cursor introduces dynamic context discovery, a pattern where AI coding agents pull relevant context on-demand rather than receiving everything upfront. This approach reduces token usage by 46.9% for MCP tools while improving response quality. Key implementations include converting long tool responses to files, using chat history during summarization, supporting Agent Skills standard, lazy-loading MCP tools, and treating terminal sessions as files. The file-based abstraction enables agents to efficiently discover and retrieve only the information they need.

Dynamic context discovery

# 1. Turning long tool responses into files

# 2. Referencing chat history during summarization

# 3. Supporting the Agent Skills open standard

# 4. Efficiently loading only the MCP tools needed

# 5. Treating all integrated terminal sessions as files