diff --git a/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchMcpService.java b/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchMcpService.java
index d11ca3f3..6dc9059e 100644
--- a/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchMcpService.java
+++ b/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchMcpService.java
@@ -2,11 +2,15 @@
 
 import com.fasterxml.jackson.databind.JsonNode;
 import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
 import io.modelcontextprotocol.client.McpSyncClient;
+import io.modelcontextprotocol.spec.McpError;
 import io.modelcontextprotocol.spec.McpSchema.CallToolRequest;
 import io.modelcontextprotocol.spec.McpSchema.CallToolResult;
 import io.modelcontextprotocol.spec.McpSchema.ListToolsResult;
 import java.util.List;
+import java.util.Locale;
 import java.util.Map;
 import lombok.RequiredArgsConstructor;
 import lombok.extern.slf4j.Slf4j;
@@ -18,11 +22,38 @@
 public class BraveSearchMcpService {
 
     private static final String BRAVE_WEB_TOOL = "brave_web_search";
+    private static final int MAX_JSON_RECURSION_DEPTH = 3;
 
     private final List<McpSyncClient> mcpClients;
-
     private final ObjectMapper objectMapper;
 
+    private static boolean looksLikeJson(String s) {
+        if (s == null) {
+            return false;
+        }
+        String t = s.trim();
+        return (!t.isEmpty()) && (t.charAt(0) == '{' || t.charAt(0) == '[');
+    }
+
+    private static String truncate(String s, int max) {
+        if (s == null) {
+            return null;
+        }
+        return s.length() > max ? s.substring(0, max) + "…" : s;
+    }
+
+    private static String getTrimmed(JsonNode obj, String field) {
+        if (obj == null) {
+            return null;
+        }
+        JsonNode n = obj.get(field);
+        if (n == null || n.isNull()) {
+            return null;
+        }
+        String v = n.asText();
+        return v == null ? null : v.trim();
+    }
+
     public JsonNode search(String query, int count, int offset) {
         McpSyncClient braveClient = resolveBraveClient();
 
@@ -33,31 +64,210 @@ public JsonNode search(String query, int count, int offset) {
 
         CallToolResult result = braveClient.callTool(request);
 
-        JsonNode content = objectMapper.valueToTree(result.content());
-        log.info("[Brave MCP Response Raw content]: {}", content.toPrettyString());
+        JsonNode raw = objectMapper.valueToTree(result.content());
+        log.info("[Brave MCP Response Raw content]: {}", raw);
 
+        ArrayNode normalized = objectMapper.createArrayNode();
+        normalizeContent(raw, normalized);
 
-        if (content != null && content.isArray()) {
-            var root = objectMapper.createObjectNode();
-            root.set("results", content);
-            return root;
-        }
+        ObjectNode root = objectMapper.createObjectNode();
+        root.set("results", normalized);
 
-        return content != null ? content : objectMapper.createObjectNode();
+        log.info("Brave 검색 결과 정규화 완료: {}건", normalized.size());
+        return root;
     }
 
+    /**
+     * MCP 클라이언트들 중 brave_web_search 툴을 가진 클라이언트 선택. 초기화되지 않은 경우 1회 initialize() 후 재시도.
+     */
     private McpSyncClient resolveBraveClient() {
         for (McpSyncClient client : mcpClients) {
-            ListToolsResult tools = client.listTools();
-            if (tools != null && tools.tools() != null) {
-                boolean found = tools.tools().stream()
-                    .anyMatch(tool -> BRAVE_WEB_TOOL.equalsIgnoreCase(tool.name()));
-                if (found) {
+            try {
+                ListToolsResult tools = client.listTools();
+                if (hasBraveTool(tools)) {
                     return client;
                 }
+            } catch (McpError e) {
+                String msg = e.getMessage() == null ? "" : e.getMessage().toLowerCase(Locale.ROOT);
+                boolean likelyUninitialized = msg.contains("not initialized")
+                    || msg.contains("uninitialized")
+                    || (msg.contains("initialize") && msg.contains("required"));
+                if (likelyUninitialized) {
+                    // 1회만 초기화 재시도
+                    try {
+                        log.warn("MCP 클라이언트 미초기화 감지 → initialize() 재시도");
+                        client.initialize();
+                        ListToolsResult tools = client.listTools();
+                        if (hasBraveTool(tools)) {
+                            return client;
+                        }
+                    } catch (Exception initError) {
+                        log.error("MCP 클라이언트 초기화 실패: {}", initError.getMessage());
+                    }
+                } else {
+                    log.debug("listTools() 예외: {}", e.getMessage());
+                }
+            } catch (Exception e) {
+                log.debug("listTools() 일반 예외: {}", e.getMessage());
             }
         }
-
         throw new IllegalStateException("Brave MCP 서버에서 brave_web_search 툴을 찾을 수 없습니다.");
     }
-}
\ No newline at end of file
+
+    private boolean hasBraveTool(ListToolsResult tools) {
+        return tools != null && tools.tools() != null &&
+            tools.tools().stream().anyMatch(t -> BRAVE_WEB_TOOL.equalsIgnoreCase(t.name()));
+    }
+
+    /**
+     * raw JSON(any shape) → results[{url,title,description}]
+     */
+    private void normalizeContent(JsonNode raw, ArrayNode out) {
+        if (raw == null || raw.isNull()) {
+            return;
+        }
+
+        if (raw.isArray()) {
+            for (JsonNode n : raw) {
+                normalizeOne(n, out);
+            }
+        } else {
+            normalizeOne(raw, out);
+        }
+    }
+
+    private void normalizeOne(JsonNode node, ArrayNode out) {
+        if (node == null || node.isNull()) {
+            return;
+        }
+
+        // 이미 {url,title,description} 형태
+        if (node.isObject() && (node.has("url") || node.has("title") || node.has("description"))) {
+            addObjectToResults((ObjectNode) node, out);
+            return;
+        }
+
+        // MCP가 주는 content item: { "type":"text", "text":"{...json...}" } 같은 형태를 방어
+        if (node.isObject() && node.has("text")) {
+            String text = node.path("text").asText("");
+            if (looksLikeJson(text)) {
+                parseJsonBlockIntoResults(text, out, 0);
+            } else {
+                // "Title: ..." 라인 포맷 등 레거시 텍스트 포맷 처리(옵션)
+                parseLegacyBlock(text, out);
+            }
+            return;
+        }
+
+        // 순수 문자열이지만 안에 JSON이 들어 있는 경우
+        if (node.isTextual() && looksLikeJson(node.asText())) {
+            parseJsonBlockIntoResults(node.asText(), out, 0);
+        }
+    }
+
+    private void parseJsonBlockIntoResults(String json, ArrayNode out, int depth) {
+        if (depth > MAX_JSON_RECURSION_DEPTH) {
+            log.warn("JSON 파싱 재귀 깊이 초과: {}", truncate(json, 100));
+            return;
+        }
+        try {
+            JsonNode parsed = objectMapper.readTree(json);
+
+            if (parsed.isArray()) {
+                for (JsonNode n : parsed) {
+                    if (n.isObject()) {
+                        addObjectToResults((ObjectNode) n, out);
+                    } else if (n.isTextual() && looksLikeJson(n.asText())) {
+                        parseJsonBlockIntoResults(n.asText(), out, depth + 1);
+                    }
+                }
+            } else if (parsed.isObject()) {
+                ObjectNode obj = (ObjectNode) parsed;
+                // 루트에 results 배열이 있는 형태 처리
+                if (obj.has("results") && obj.get("results").isArray()) {
+                    for (JsonNode n : obj.get("results")) {
+                        if (n.isObject()) {
+                            addObjectToResults((ObjectNode) n, out);
+                        } else if (n.isTextual() && looksLikeJson(n.asText())) {
+                            parseJsonBlockIntoResults(n.asText(), out, depth + 1);
+                        }
+                    }
+                    return;
+                }
+                if (obj.has("text") && obj.get("text").isTextual() && looksLikeJson(
+                    obj.get("text").asText())) {
+                    // {text:"{...}"} 같은 한 번 더 감싼 케이스
+                    parseJsonBlockIntoResults(obj.get("text").asText(), out, depth + 1);
+                } else {
+                    addObjectToResults(obj, out);
+                }
+            } else if (parsed.isTextual() && looksLikeJson(parsed.asText())) {
+                parseJsonBlockIntoResults(parsed.asText(), out, depth + 1);
+            }
+        } catch (Exception e) {
+            log.warn("Brave MCP JSON 파싱 실패: {}\n원인: {}", truncate(json, 400), e.getMessage());
+        }
+    }
+
+    private void addObjectToResults(ObjectNode obj, ArrayNode out) {
+        String url = getTrimmed(obj, "url");
+        String title = getTrimmed(obj, "title");
+        String desc = getTrimmed(obj, "description");
+
+        // 세 필드 중 하나라도 있으면 결과로 채택
+        if ((url != null && !url.isBlank()) ||
+            (title != null && !title.isBlank()) ||
+            (desc != null && !desc.isBlank())) {
+
+            ObjectNode one = objectMapper.createObjectNode();
+            if (url != null) {
+                one.put("url", url);
+            }
+            if (title != null) {
+                one.put("title", title);
+            }
+            if (desc != null) {
+                one.put("description", desc);
+            }
+            out.add(one);
+        }
+    }
+
+    // 레거시 "Title:..., URL:..., 본문..." 형태의 텍스트 파서(있으면 도움, 없어도 무방)
+    private void parseLegacyBlock(String text, ArrayNode out) {
+        if (text == null || text.isBlank()) {
+            return;
+        }
+
+        String[] lines = text.split("\\r?\\n");
+        String title = null, url = null;
+        StringBuilder body = new StringBuilder();
+
+        for (String line : lines) {
+            String trimmed = line.trim();
+            if (trimmed.regionMatches(true, 0, "Title:", 0, 6)) {
+                if (title != null && url != null && body.length() > 0) {
+                    ObjectNode one = objectMapper.createObjectNode();
+                    one.put("title", title);
+                    one.put("url", url);
+                    one.put("description", body.toString().trim());
+                    out.add(one);
+                    body.setLength(0);
+                }
+                title = trimmed.substring(6).trim();
+            } else if (trimmed.regionMatches(true, 0, "URL:", 0, 4)) {
+                url = trimmed.substring(4).trim();
+            } else {
+                body.append(line).append('\n');
+            }
+        }
+
+        if (title != null && url != null) {
+            ObjectNode one = objectMapper.createObjectNode();
+            one.put("title", title);
+            one.put("url", url);
+            one.put("description", body.toString().trim());
+            out.add(one);
+        }
+    }
+}
diff --git a/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchRagService.java b/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchRagService.java
index 5a468cdb..d29bd35b 100644
--- a/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchRagService.java
+++ b/cs25-service/src/main/java/com/example/cs25service/domain/ai/service/BraveSearchRagService.java
@@ -1,6 +1,7 @@
 package com.example.cs25service.domain.ai.service;
 
 import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
 import java.util.ArrayList;
 import java.util.List;
 import java.util.Map;
@@ -13,54 +14,126 @@
 @RequiredArgsConstructor
 public class BraveSearchRagService {
 
+    private final ObjectMapper objectMapper = new ObjectMapper();
+
     public List<Document> toDocuments(Optional<JsonNode> resultsNodeOpt) {
         List<Document> documents = new ArrayList<>();
+        resultsNodeOpt.ifPresent(root -> {
+            for (JsonNode r : root.path("results")) {
+                // 1) 표준 필드 우선 사용
+                String url = getText(r, "url");
+                String title = getText(r, "title");
+                String description = getText(r, "description");
 
-        resultsNodeOpt.ifPresent(resultsNode -> {
-            resultsNode.path("results").forEach(result -> {
-                String text = result.path("text").asText("");
-                if (text.isBlank()) {
-                    return;
+                // 2) text 필드가 JSON 문자열일 수 있음
+                if (isBlank(url) && isBlank(title) && isBlank(description)) {
+                    String text = getText(r, "text");
+                    if (!isBlank(text) && looksLikeJson(text)) {
+                        try {
+                            JsonNode inner = objectMapper.readTree(text);
+                            url = isBlank(url) ? getText(inner, "url") : url;
+                            title = isBlank(title) ? getText(inner, "title") : title;
+                            description =
+                                isBlank(description) ? getText(inner, "description") : description;
+                        } catch (Exception ignored) { /* fall back below */ }
+                    }
                 }
 
-                // 여러 문서가 한 개의 텍스트에 포함되어 있으므로 줄 단위로 분리
-                String[] lines = text.split("\\n");
-
-                String title = null;
-                String url = null;
-                StringBuilder contentBuilder = new StringBuilder();
-
-                for (String line : lines) {
-                    if (line.startsWith("Title:")) {
-                        if (title != null && url != null && contentBuilder.length() > 0) {
-                            // 이전 문서를 저장
-                            documents.add(new Document(
-                                title,
-                                contentBuilder.toString().trim(),
-                                Map.of("title", title, "url", url)
-                            ));
-                            contentBuilder.setLength(0);
+                // 3) 레거시 포맷: "Title:/URL:" 라인 파싱
+                if (isBlank(url) && isBlank(title) && isBlank(description)) {
+                    String text = getText(r, "text");
+                    if (!isBlank(text)) {
+                        ParsedLegacy pl = parseLegacyBlock(text);
+                        if (pl != null) {
+                            url = pl.url != null ? pl.url : url;
+                            title = pl.title != null ? pl.title : title;
+                            description = pl.body != null ? pl.body : description;
                         }
-                        title = line.replaceFirst("Title:", "").trim();
-                    } else if (line.startsWith("URL:")) {
-                        url = line.replaceFirst("URL:", "").trim();
-                    } else {
-                        contentBuilder.append(line).append("\n");
                     }
                 }
 
-                // 마지막 문서 저장
-                if (title != null && url != null && contentBuilder.length() > 0) {
-                    documents.add(new Document(
-                        title,
-                        contentBuilder.toString().trim(),
-                        Map.of("title", title, "url", url)
-                    ));
+                // 4) 아무 것도 없으면 스킵
+                if (isBlank(url) && isBlank(title) && isBlank(description)) {
+                    continue;
                 }
-            });
-        });
 
+                // 5) Document id는 URL > title 우선
+                String id = !isBlank(url) ? url : title;
+                String body = !isBlank(description) ? description
+                    : (!isBlank(title) ? title : (url != null ? url : ""));
+
+                documents.add(new Document(
+                    id,
+                    body,
+                    Map.of(
+                        "title", title == null ? "" : title,
+                        "url", url == null ? "" : url
+                    )
+                ));
+            }
+        });
         return documents;
     }
 
+    /* ----------------- helpers ----------------- */
+
+    private String getText(JsonNode n, String field) {
+        return (n != null && n.has(field) && n.get(field).isTextual())
+            ? n.get(field).asText().trim() : null;
+    }
+
+    private boolean isBlank(String s) {
+        return s == null || s.isBlank();
+    }
+
+    private boolean looksLikeJson(String s) {
+        String t = s.trim();
+        return (t.startsWith("{") && t.endsWith("}")) ||
+            (t.startsWith("[") && t.endsWith("]")) ||
+            t.contains("\"url\":") || t.contains("\"title\":") || t.contains("\"description\":");
+    }
+
+    /**
+     * "Title: ..." / "URL: ..." 블록을 관대한 방식으로 파싱
+     */
+    private ParsedLegacy parseLegacyBlock(String text) {
+        if (text == null) {
+            return null;
+        }
+        String[] lines = text.split("\\r?\\n");
+        String title = null, url = null;
+        StringBuilder body = new StringBuilder();
+
+        // "Title:" / "URL:" 키워드는 대소문자 무시 + 앞뒤 공백 관대 처리
+        for (String raw : lines) {
+            String line = raw == null ? "" : raw.trim();
+            if (line.regionMatches(true, 0, "Title:", 0, 6)) {
+                // 이전 누적 flush
+                // (여기서는 단일 레코드만 기대하므로 flush 없이 교체)
+                title = line.substring(6).trim();
+            } else if (line.regionMatches(true, 0, "URL:", 0, 4)) {
+                url = line.substring(4).trim();
+            } else {
+                body.append(line).append('\n');
+            }
+        }
+        String desc = body.toString().trim();
+        if (isBlank(title) && isBlank(url) && isBlank(desc)) {
+            return null;
+        }
+        return new ParsedLegacy(title, url, desc);
+    }
+
+    private static class ParsedLegacy {
+
+        final String title;
+        final String url;
+        final String body;
+
+        ParsedLegacy(String title, String url, String body) {
+            this.title = title;
+            this.url = url;
+            this.body = body;
+        }
+    }
 }