diff --git a/plugins/web/keyless_mcp.py b/plugins/web/keyless_mcp.py
index 1855c8897c..116feaff32 100644
--- a/plugins/web/keyless_mcp.py
+++ b/plugins/web/keyless_mcp.py
@@ -120,7 +120,12 @@ def use_keyless(name: str, api_key: str) -> bool:
 def _parse_mcp_body(body: str) -> str:
     """First text content item from an MCP tools/call response — plain-JSON bodies
     (Parallel) or SSE ``data: {...}`` lines (Exa). Raises :class:`KeylessMCPError` for
-    JSON-RPC errors and ``isError`` tool results (e.g. Exa's free-tier rate limit)."""
+    JSON-RPC errors and ``isError`` tool results (e.g. Exa's free-tier rate limit).
+
+    SSE frames are split on ``\\n`` only: ``str.splitlines()`` also breaks on U+0085 /
+    U+2028 / U+2029, which legitimately occur inside CJK page text and would cut a
+    ``data:`` line in two. A parsed envelope with no text is reported as such — it is
+    the vendor's answer, not an unrecognized shape."""
 
     def _from_payload(payload: str) -> Optional[str]:
         payload = payload.strip()
@@ -134,19 +139,21 @@ def _parse_mcp_body(body: str) -> str:
         texts = [c.get("text", "") for c in result.get("content") or [] if isinstance(c, dict)]
         if result.get("isError"):
             raise KeylessMCPError(" ".join(t for t in texts if t) or "MCP tool call failed")
-        return next((str(t) for t in texts if t), None)
+        return next((str(t) for t in texts if t), "")
 
     stripped = body.strip()
     candidates = [stripped] if stripped.startswith("{") else []
-    candidates += [line[len("data: "):] for line in body.splitlines() if line.startswith("data: ")]
+    candidates += [line[len("data: "):] for line in body.split("\n") if line.startswith("data: ")]
+    envelope_seen = False
     for candidate in candidates:
         try:
             text = _from_payload(candidate)
         except json.JSONDecodeError:
             continue
-        if text is not None:
+        if text:
             return text
-    raise KeylessMCPError("Unrecognized MCP response shape")
+        envelope_seen = envelope_seen or text == ""
+    raise KeylessMCPError("MCP response contained no text content" if envelope_seen else "Unrecognized MCP response shape")
 
 
 def mcp_call(url: str, tool: str, arguments: Dict[str, Any], timeout: int = _TIMEOUT_SECONDS) -> str:
@@ -161,7 +168,10 @@ def mcp_call(url: str, tool: str, arguments: Dict[str, Any], timeout: int = _TIM
         raise KeylessMCPError(f"request failed: {exc}") from exc
     if response.status_code >= 400:
         raise KeylessMCPError(f"HTTP {response.status_code}: {response.text[:300]}")
-    return _parse_mcp_body(response.text)
+    # JSON-RPC and SSE bodies are UTF-8 by spec, but ``text/event-stream`` carries no charset and
+    # ``requests`` then decodes ``.text`` as ISO-8859-1 — mojibake for every non-ASCII result and,
+    # for CJK, stray U+0085 line breaks that made the envelope unparseable.
+    return _parse_mcp_body(response.content.decode("utf-8", errors="replace"))
 
 
 # --- Parallel (search.parallel.ai) — JSON text payloads -----------------------
diff --git a/tests/tools/test_web_keyless_fallback.py b/tests/tools/test_web_keyless_fallback.py
index 1de9fba14b..762d3c52f7 100644
--- a/tests/tools/test_web_keyless_fallback.py
+++ b/tests/tools/test_web_keyless_fallback.py
@@ -89,6 +89,29 @@ class TestParseMcpBody:
         with pytest.raises(keyless_mcp.KeylessMCPError):
             keyless_mcp._parse_mcp_body("<html>nope</html>")
 
+    def test_cjk_sse_body_survives_charset_less_event_stream(self):
+        """Exa answers ``text/event-stream`` without a charset; ``requests`` then decodes ``.text`` as
+        ISO-8859-1, and the U+0085 inside CJK UTF-8 sequences split the ``data:`` line under
+        ``splitlines()`` — a valid CJK result surfaced as "Unrecognized MCP response shape"."""
+        import requests
+
+        title = "光伏发电站组件清洗与性能监测规范"
+        payload = {"result": {"content": [{"type": "text", "text": f"Title: {title}\nURL: https://x.example"}]}}
+        response = requests.Response()
+        response.status_code = 200
+        response.headers["Content-Type"] = "text/event-stream"
+        response.encoding = "ISO-8859-1"  # what the adapter picks for text/* without a charset
+        response._content = f"event: message\ndata: {json.dumps(payload, ensure_ascii=False)}\n\n".encode("utf-8")
+        assert "\x85" in response.text  # the vendor body really does decode to mojibake via .text
+        with patch.object(requests, "post", return_value=response):
+            text = keyless_mcp.mcp_call(keyless_mcp.EXA_MCP_URL, "web_search_exa", {"query": title})
+        assert title in text
+
+    def test_parsed_envelope_without_text_names_the_condition(self):
+        body = json.dumps({"result": {"content": []}})
+        with pytest.raises(keyless_mcp.KeylessMCPError, match="no text content"):
+            keyless_mcp._parse_mcp_body(body)
+
 
 class TestExaTextParsing:
     def test_parses_blocks(self):