Source: Discord Chinese channel message https://discord.com/channels/1409545250376122451/1409545253748473931/1536223620886437948
A user hit a ChatGPT/custom MCP tool-result size error when calling memory_search with limit=10. The screenshot showed the result payload at about 66k characters in one line before the host rejected it.
Root cause: memory_search was returning full memory content for each hit. That makes search behave like a bulk export path and conflicts with the intended progressive flow: search should return compact previews plus stable memory_id / nowledgemem:// handles, then the agent should read only the specific memory it actually needs in full.
Expected behavior:
memory_search returns bounded previews and stable read handles.- Long memory content is marked as truncated instead of overflowing the host result limit.
- Agents use
get_memory_by_id(memory_id) or the resource URI for full content only when needed.
A fix is prepared for the next build. Keep this open until the build containing the bounded search payload is available to users.
Source: Discord Chinese channel message https://discord.com/channels/1409545250376122451/1409545253748473931/1536223620886437948
A user hit a ChatGPT/custom MCP tool-result size error when calling
memory_searchwithlimit=10. The screenshot showed the result payload at about 66k characters in one line before the host rejected it.Root cause:
memory_searchwas returning full memory content for each hit. That makes search behave like a bulk export path and conflicts with the intended progressive flow: search should return compact previews plus stablememory_id/nowledgemem://handles, then the agent should read only the specific memory it actually needs in full.Expected behavior:
memory_searchreturns bounded previews and stable read handles.get_memory_by_id(memory_id)or the resource URI for full content only when needed.A fix is prepared for the next build. Keep this open until the build containing the bounded search payload is available to users.