Memory (reference) vs context-mode
Both are memory & reasoning MCP servers, and we ran both through the same test (recall across sessions) in each app. Memory (reference) passed 4 of 4 apps we tested it in; context-mode passed 4 of 4.
Test results side by side
| Memory (reference) | context-mode | |
|---|---|---|
| Claude Code | Passed 2026-10-19 | Passed 2026-10-13 |
| Cursor | Passed 2026-10-19 | Passed 2026-10-13 |
| Claude Desktop | Passed 2026-10-21 | Passed 2026-10-14 |
| ChatGPT (Codex) | Passed 2026-10-20 | Passed 2026-10-14 |
| Tools exposed (protocol check) | 9 tools | 11 tools |
| GitHub stars | 91,266 (shared repo) | 29,079 |
| License | NOASSERTION | NOASSERTION |
| Last push | 2026-10-13 | 2026-10-14 |
Results from our own harness runs, dated per cell. Stars and last push from GitHub at our last data refresh. We don’t pick a winner here: which one fits depends on your app and setup.
What each does
Memory (reference): A simple knowledge-graph memory the agent can add to and search across sessions, stored in a local JSON file.
context-mode: Keeps large tool outputs out of the context window by summarizing/indexing them.
