Skip to content
Nathan GalloSoftware engineering · AI applications
All projects
Open sourceCode IntelligenceCLIModified ForkTree-sitter

Codebase Memory CLI

A command-line fork of DeusData/codebase-memory-mcp. The indexing and search engine is DeusData's work. My fork runs it without an MCP server.

SourceDocumentationAttribution to the original project
Terminal output: codebase-memory-cli indexes the public Agent-Workflow repository, 6,938 nodes and 20,726 edges, then lists the four callers of _accepted_run
Indexing the public Agent-Workflow repo, then tracing callers

Summary

What it does. The engine comes from DeusData/codebase-memory-mcp, including code parsing, the code graph, indexing, storage, the file watcher and language support.

My part. I removed the supported MCP server, made the command line the entry point, adapted the JSON output and the install and security behavior for agents, and tested the fork. DeusData wrote the underlying architecture and code-intelligence engine.

System diagrams

Simplified drawings of how the parts fit together. They show the design, not a record of real runs.

DeusData's indexing engine, my command-line forkDeusData's indexing engine, my command-line fork
View full-size diagram View full-size diagram Diagram source file

Based on the public fork's README and UPSTREAM_ATTRIBUTION.md. DeusData wrote the parser and indexing engine.

Show 1 more diagrams

System diagrams

These diagrams show the architecture, who is responsible for what, and where evidence comes from, based on the project record. They describe the implementation and its stated limits. They are not independent test results.

01 / Who controls what

Who built which part

DeusData · inherited engine

  • Tree-sitter parsing and code graph
  • Storage, indexing, daemon and watcher foundation
  • Language and relationship discovery

Modified fork · CLI boundary

  • Supported CLI and stable JSON for agents
  • MCP server removed from the supported features
  • Changed install, hook and security behavior

Reader · checks the evidence

  • Use graph results to find relevant code
  • Check the exact source revision and the coverage
  • A missing index hit does not prove something is absent

How it works

  • Original engine and architecture inherited from DeusData/codebase-memory-mcp
  • Command-line output for people and stable JSON output for agents and scripts
  • Install and hook behavior changed so using the CLI does not quietly register an MCP server

How it's built

  • The daemon, workers, watcher, storage and indexing engine, and much of the runtime are original work by DeusData.

Rule-based control

  • Limits on which folders can be indexed
  • Locks that stop two changes to one project at once, and checks that versions match

Where AI models are used

  • Indexing and queries do not use a model.
  • Agents call the CLI and decide how to use what it returns.

Failure and recovery

  • The tool reports index coverage before anyone claims something is absent or complete.
  • Hooks are optional. If a hook fails, it does not block the CLI.
  • A Windows failure showed up late in testing. A database cleanup step was missing. It was fixed and the corrected version was tested again before merge.

Measured evidence

Cross-platform integration acceptance

measured

34 GitHub checks passed · 91 affected Jenkins checks · Windows integration: 28/28

Before merge, the final release-tooling revision passed all GitHub checks, the Jenkins production build and focused integration suites, and both Windows shards.

Integration testing of the merged tree only. It does not cover release or package publication.

2026-10-03source ea4335f67060
method and source

Evidence & documentation

Original project: DeusData/codebase-memory-mcpFork attributionCross-platform integration acceptance

Visitor analytics: page interactions, source category, browser information, and network IP metadata are recorded in private server logs. No analytics cookies are used. Opt out of analytics in this browser.