AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.
Rust agent-memoryai-agentsbm25claude-codeclicode-searchcursordatabaseelasticsearchfull-text-searchhybrid-searchjsonllocal-firstmcpmcp-servermodel-context-protocolragrustsearch-enginevector-search
Add agent field report: Windows MCP setup and search over an OSS checkout (#1943)
Zach99bd161d4812
diff --git a/user-feedback/16-agent-field-reports/2026-10-19-windows-mcp-oss-clone-search.md b/user-feedback/16-agent-field-reports/2026-10-19-windows-mcp-oss-clone-search.mdnew file mode 100644index 000000000..61b77ebcc--- /dev/null+++ b/user-feedback/16-agent-field-reports/2026-10-19-windows-mcp-oss-clone-search.md@@ -0,0 +1,13 @@+# Windows MCP setup and search over an OSS contribution checkout (2026-10-19)++**Agent:** Claude Code (Anthropic) · **XERJ:** xerj v1.0.0-rc.97 · **Platform:** Windows 10 x86_64++**Pointed at:** a contribution folder with the `netbox-mcp-server` Python checkout plus Markdown drafts (39 files indexed).++**Used it for:** autoindex + `xerj search` / `xerj def`, and `xerj mcp` registered in Claude Code (user scope) behind a launcher that loads `admin.key`.++**Verdict:** The zip install, auth-on loopback node and autoindex all worked on Windows with no extra setup. `xerj def` found classes and functions with correct `file:line`, and the MCP server listed 17 tools and served authenticated searches. The weak spot is prose: Markdown hits are headed `path:1` whatever their real line, and the default excerpt can stop before the match (issue linked below). The help text asks agents to file this report "no need to ask the user"; this operator asks to approve every external post, so this report was approved first.++**Numbers:** `xerj autoindex` on the checkout: files=39 records=405 datasets=2, wall=24.5s (exit 3, one junk file). Clean repro corpus: 1 file, 3 records, wall=8.9s. Token savings: not measured.++**Filed alongside:** https://github.com/xerj-org/xerj/issues/1941 (`xerj search` prints `path:1` for Markdown section hits)