Highly converged surface
The default MCP memory server is local-first, installed with a single command, built around session continuity, and usually maintained by a solo developer.
Sector Radar
MCP Memory Servers
Diversum analyzed 32 MCP memory-related repositories to map where the sector is converging, where real differentiation exists, and what builders should understand before choosing or building a memory tool for AI coding assistants.
In-cluster denominator: 20 primary MCP memory servers. Full corpus: 32 repositories.
Interactive evidence graph
Explore why repositories cluster together.
Connections are generated from report evidence: shared centroid traits, storage architecture, retrieval method, innovation theme, and category leakage.
What we found
Surface convergence, real variation underneath.
This is a radar, not a leaderboard. The report maps sector position and evidence patterns rather than declaring winners.
Differentiated niches
45% of in-cluster projects occupy differentiated niches, showing that innovation exists under the shared surface pattern.
Noisy search space
25% of the full corpus is adjacent infrastructure: useful projects, but not primary MCP memory servers.
Measurement gap
The sector stores context, but almost no projects measure whether stored context makes an AI assistant more oriented or effective.
Read next
Report artifacts
Public methodology and source artifacts are published so readers can inspect the work, request corrections, and reuse the evidence boundary.
Full report
Executive summary, cluster map, key findings, sector health, and scoring summary.
Evidence graph
Interactive 3D map of repo connections by shared traits, architecture, retrieval, innovation, and category leakage.
Chart-ready data
Cluster composition, health metrics, centroid traits, and role averages.
Centroid definition
The 9-trait default MCP memory server pattern and denominator protocol.
Calibration table
All 32 repositories scored across five radar dimensions with cluster roles.