# CodeNib [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/sysevol-ai/CodeNib  
**GitHub Stars:** 107  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/codenib

## Description
Ranked codebase search and static symbol navigation for coding agents.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "codenib": {
    "command": "uvx",
    "args": ["codenib"]
  }
}
```

## Documentation

## What CodeNib does

CodeNib gives coding agents structured ways to locate and inspect source code in a local repository. It combines ranked retrieval with a typed symbol graph, allowing an agent to start from a code question, receive source anchors, and then follow definitions or static references. Results include file and line references so the agent can inspect the underlying evidence before describing an implementation or call path.

The CodeGraph workflow supports graph backends for Python, Go, Rust, C and C++, C#, Java, Ruby, PHP, Kotlin, Scala, JavaScript, and TypeScript. Swift and Lua support chunking and retrieval. Language coverage, provider prerequisites, and graph behavior vary by language.

## How it works

CodeNib builds local search views and an indexed symbol graph from a clean repository. `explore_context` provides ranked, bounded context for a question or investigation. `dependency_subgraph` follows incoming and outgoing static references, with a single request able to traverse relationships up to depth 8. Graph relationships are resolved through SCIP or language-server indexers and can represent calls, fields, type uses, and function values.

The graph describes statically indexed relationships rather than guaranteed runtime behavior. Dynamic or external relationships may be absent, and a reference that looks like a call still requires source inspection to establish how it is used. Native agent search and file reads remain useful for known strings or files.

## Setup and configuration

Install the graph and MCP components with Python 3.10 or newer:

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

The setup requires Git, a clean repository, and an MCP-capable agent such as Claude Code or Codex. After initialization, ask the agent to use `explore_context` or another CodeNib tool for a repository investigation. CodeNib itself does not require a model, API key, or GPU; the connected agent supplies the model.

Indexing and rebuild costs depend on the repository and language tooling. The product reuses or rebuilds views, while file-level delta repair is not enabled. An optional grep-to-Jev route can use OpenRouter for planning and reranking; selected code is sent to remote models, and authorization previews are opt-in.

## Tools and capabilities

- Retrieve ranked, bounded source context with file and line citations through `explore_context`.
- Navigate definitions and statically indexed references across symbols.
- Trace multi-hop dependency relationships with `dependency_subgraph`.
- Use one retrieval and navigation workflow across supported programming languages.
- Export a cited local Wiki to GitHub Pages or browse generated local and public examples.

## Limitations and notes

CodeNib is in beta, so public interfaces may change. Graph results reflect indexed static relationships and can miss dynamic behavior, external dependencies, or relationships not resolved by the configured language backend. Coverage and setup requirements differ between languages. The optional remote reranking route has different privacy and model-use characteristics from the local, model-free CodeGraph workflow.

_Full upstream README: https://allmcps.com/mcp/codenib/readme_

