# google-surf-mcp

An MCP server that stores the results of web searches, academic papers and code exploration in a local, project-scoped knowledge graph, so an AI agent's research accumulates instead of repeating the same lookups.

google-surf-mcp gives an AI agent search, extraction and memory tools over MCP: Google web search by default with an optional SearchApi provider, Google Scholar for paper metadata, and HTML or PDF extraction that returns titles, authors, DOIs and body text. Results, along with the session's intent, plans, experiments and decisions, are written into a local knowledge graph scoped to the current project.

Retrieval combines exact match, BM25, vector search, code and graph search and live web results, fusing them with reciprocal rank fusion and a shared reranker. Projects stay isolated from each other by default, and only explicit, verifiable links, such as a matching DOI, repository URL or declared alias, connect knowledge across projects. Seven tools are exposed by default, and a research mode that opens the local database and graph can be disabled to use only search and extraction.

Website: https://github.com/HarimxChoi/google-surf-mcp