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
docs(field-report): record MCP budget identity validation (#2265)
Jarvis5ee3ac5cd677
diff --git a/user-feedback/16-agent-field-reports/2026-10-23-mcp-budget-identity.md b/user-feedback/16-agent-field-reports/2026-10-23-mcp-budget-identity.mdnew file mode 100644index 000000000..e3483dd8f--- /dev/null+++ b/user-feedback/16-agent-field-reports/2026-10-23-mcp-budget-identity.md@@ -0,0 +1,13 @@+# MCP budget identity regression validation (2026-10-23)++**Agent:** OpenAI Codex · **XERJ:** xerj v1.0.0-rc.98 · **Platform:** macOS arm64++**Pointed at:** A fresh temporary node with seeded documents shaped like autoindexed data rows and same-path files in separate indices; no user corpus was indexed.++**Used it for:** Validating MCP JSON search budget deduplication through HTTP fixtures and the release MCP executable against a real node.++**Verdict:** A generous max_tokens budget could silently remove independent rows or same-path hits from different corpora. Qualifying spans by source and record identity fixes the reproduced cases. The plain-string source projection also needed provenance fields. This is adapter correctness validation, not a retrieval-quality or performance evaluation.++**Numbers:** MCP crate suite: 133 passed, 0 failed. Release-MCP/real-node checks: four plain/DSL and row/cross-index combinations retained both hits, with and without a budget.++**Filed alongside:** [PR #2255](https://github.com/xerj-org/xerj/pull/2255).