Breaking

DeusData’s codebase-memory-mcp v0.11.0 targets faster, leaner code context for AI agents

September 21, 2026 at 17:51 EDT

Software Dev & Coding
AI Agents
Open Source

DeusData has released codebase-memory-mcp v0.11.0, sharpening an open-source code intelligence engine that turns repositories into persistent knowledge graphs for AI coding agents. The update, published on September 15, 2026, focuses on reliability, memory control and more compact outputs as developers look for ways to stop agents from repeatedly reading entire codebases.

Sep. 15, 2026 · DeusData

Codebase context becomes a graph

codebase-memory-mcp v0.11.0 hardens a local knowledge-graph engine for AI coding agents, emphasizing faster structural lookup, lower token use and more predictable indexing on large repositories.

162

documented language coverage

MCP tools

<1 ms

structural query latency

agent client surfaces

Token load per five structural queries

The graph path is reported at about 3,400 tokens versus about 412,000 for file-by-file exploration.

412,000file-by-file tokens

3,400graph-query tokens

Takeaway: the reported workload cuts token use by 99.2% when agents query the graph first.

Agent workflow

Parse repository

tree-sitter and Hybrid LSP extract structure.

→

Persist graph

SQLite stores functions, classes, calls and routes.

→

Answer tools

Agents ask targeted structural questions.

Why developers like it

Faster orientation in unfamiliar repositories
Fewer full-file reads for routine exploration
Local processing for private codebases

What still needs care

Graph answers can become pointer-chasing
Static analysis has semantic limits
Shared artifacts need repo hygiene

v0.11.0 is a reliability release

The release merged 171 pull requests, tightens memory-budget behavior, compacts default outputs and requires one full reindex after upgrading before incremental indexing resumes.

171 PRsmemory budget enforcedone rebuild after upgrade

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.