← Back to archive

Repo of the Day

sysevol-ai/CodeNib: Find source context and trace calls across languages in Claude Code, Codex and other MCP agents.

Published: Nov 18, 2026

Open repository ↗

Find source context and trace calls across languages in Claude Code, Codex and other MCP agents. - sysevol-ai/CodeNib

Summary

CodeNib indexes a repository into local search and a typed symbol graph, then uses that index to build a browsable Wiki with call paths and file-and-line citations. Through MCP and plugins for Claude Code, Codex, Grok Build, and DeepSeek Harness, it gives coding agents source-backed definitions, callers, and dependency paths.

What it is useful for

CodeNib is useful when engineers need to understand unfamiliar or large codebases, trace where behavior is implemented, or determine which public entry points reach a function. The README’s Requests example follows Session.send() through resolve_redirects() to get_redirect_target(); its Gin example traces GET() through router registration to node.insertChild(). The Wiki groups code into areas, lets readers follow a recorded path, and connects each explanation to the source. “Ask this repository” answers questions with citations using a configured model.

Two previews cover change-related work: codenib guard lists tests a change can reach, currently for Python, while codenib threads reads a GitHub issue or pull request once and packages the request, review comments, failing checks, linked threads, and referenced code for an agent.

CodeNib registers 14 languages. Python, Go, Rust, C/C++, JavaScript, and TypeScript use an incremental LSP or clangd backend; C#, Java, Ruby, PHP, Kotlin, and Scala have code graphs; Swift and Lua are search-only. Indexed external or dynamic relationships may be missing, so engineers should inspect source to confirm actual calls. Indexing and rebuild costs apply, file-level delta repair is not enabled, and CodeNib is beta, so public interfaces may change.

How engineers can use it

For agent-assisted context, install the graph and MCP extras, then initialize the target checkout:

python -m pip install "codenib[graph,mcp]"
codenib codegraph init /path/to/your/repo

This requires Python 3.10+, Git, and Claude Code or Codex. It builds local search and the symbol graph, connects installed agent clients, accepts uncommitted edits, and preserves the checkout. A practical first prompt is:

Use CodeNib’s explore_context to find where request retry behavior is implemented. Cite the files and lines.

To build a local Wiki:

python -m pip install "codenib[graph]"
codenib wiki /path/to/your/repo --preset graph

It opens at http://localhost:3000; add --generate --model <provider/model> for model-written pages. For issue or pull-request context, run codenib threads context 812 from the repository. Check codenib toolchain status to see which language backends are available locally.