fix(fable-mcp): fix SSE parser and add stream retry for race condition

- Parse multi-line SSE format correctly (event: on separate line, chunk in data.chunk)
- Retry stream GET up to 10x with 300ms backoff when 404 (buffer not ready yet)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

fable-mcp/fable_mcp/tools/chat.py

@@ -1,10 +1,11 @@
 """MCP tools for Fable chat — create conversations and stream responses."""
 from __future__ import annotations
 
+import asyncio
 import json
 from typing import Any
 
-from fable_mcp.client import FableClient
+from fable_mcp.client import FableClient, FableAPIError
 
 
 async def list_conversations(
@@ -28,8 +29,12 @@ async def send_message(
     """Send a message to Fable and return the full assistant response.
 
     Creates a new MCP conversation if conversation_id is None.
-    Posts the user message to start generation, then streams the SSE buffer
+    Posts the user message to start generation, then streams the SSE buffer.
-    and accumulates token chunks into a single string.
+
+    SSE event format:
+        id: <N>
+        event: <type>        # chunk | done | tool_call | status | ...
+        data: <json>
 
     Returns:
         Dict with:
@@ -54,21 +59,34 @@ async def send_message(
     tool_calls: list[Any] = []
 
     stream_path = f"/api/chat/conversations/{conversation_id}/generation/stream"
-    async for raw_line in client.stream_get(stream_path):
+
-        if not raw_line.startswith("data: "):
+    # Retry connecting to the stream briefly — the background task may not have
-            continue
+    # created the generation buffer by the time we issue the GET.
-        payload_str = raw_line[len("data: "):]
+    for attempt in range(10):
         try:
-            event = json.loads(payload_str)
+            event_type: str | None = None
-        except json.JSONDecodeError:
+            async for raw_line in client.stream_get(stream_path):
-            continue
+                if raw_line.startswith("event: "):
-        event_type = event.get("type")
+                    event_type = raw_line[len("event: "):].strip()
-        if event_type == "token":
+                elif raw_line.startswith("data: "):
-            tokens.append(event.get("content", ""))
+                    payload_str = raw_line[len("data: "):]
-        elif event_type == "tool_call":
+                    try:
-            tool_calls.append(event)
+                        data = json.loads(payload_str)
-        elif event_type == "done":
+                    except json.JSONDecodeError:
-            break
+                        continue
+                    if event_type == "chunk":
+                        tokens.append(data.get("chunk", ""))
+                    elif event_type == "tool_call":
+                        tool_calls.append(data.get("tool_call", data))
+                    elif event_type == "done":
+                        break
+                    event_type = None  # reset after consuming data line
+            break  # stream completed successfully
+        except FableAPIError as exc:
+            if exc.status_code == 404 and attempt < 9:
+                await asyncio.sleep(0.3)
+                continue
+            raise
 
     return {
         "conversation_id": conversation_id,

fable-mcp/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "fable-mcp"
-version = "0.2.2"
+version = "0.2.3"
 description = "MCP server for Fabled Assistant"
 requires-python = ">=3.12"
 dependencies = [