CodeNib MCP
CodeNib indexes a repository and gives Claude Code and Codex ranked context with files and lines plus a graph of call chains
- 91 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Indexes the whole repository and connects its MCP tools to installed agents
- The optional grep and Jev route sends selected code to remote models through OpenRouter
Install
- Works in
Manual install
python -m pip install "codenib[graph,mcp]"Requires Python 3.10 or newer, Git, and Claude Code or Codex.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Indexes the whole repository and connects its MCP tools to installed agents
- The optional grep and Jev route sends selected code to remote models through OpenRouter
This is third-party code. Review the repository files before installing.
What it does
CodeNib builds a local search and a typed symbol graph for a repository, then connects them to installed agents as MCP tools. The explore_context tool takes a plain-language query and returns ranked code fragments with file and line references. The dependency_subgraph tool collects a chain of calling functions up to eight steps deep in one call. SCIP and LSP indexers resolve the reference graph, so links are determined by the compiler. Graph backends exist for Python, Go, Rust, C and C++, C#, Java, Ruby, PHP, Kotlin, Scala, JavaScript and TypeScript, while Swift and Lua support chunking and retrieval only. CodeNib itself needs no model, API key or GPU: your agent's model does the reasoning. The project can also export a wiki of the code with citations to GitHub Pages and contains an optional route with grep planning through OpenRouter, but the main path runs entirely locally.
Who it is for. Teams with large multi-language codebases where Claude Code or Codex spend many steps finding the right place and call chain.
Good fit when
- You need to know where request retries are implemented in a project and get an answer with files and lines
- Before changing a public function you want to see the four-call chain that leads to it
- The repository is written in Go, Rust and TypeScript, and text search returns too much noise
- A new team member needs a code wiki with citations on your own GitHub Pages
- You want to compare how an agent finds context through grep and through the symbol graph
Not a fit when
- You search for a known string or file: the agent's native search is quicker and indexing takes time
- The repository has uncommitted changes: the instructions require a clean state
- The project uses a language without a graph backend and you need call links from the graph
- Python 3.10 or newer is not installed and cannot be installed
Example request
Use CodeNib's explore_context to find where request retry behavior is implemented, and cite the files and linesMCP
- Transport
- stdio
- Authentication
- not required
Limitations
- The project is in beta, and public interfaces may change before a stable release.
- To start you need Python 3.10 or newer, Git, a clean repository and Claude Code or Codex; language toolchains differ.
- Dynamic and external relations may be missing from the graph, and which references actually call a function is established by reading the source.
- In version 0.2.4 the optional grep and Jev route uses OpenRouter, and the selected code goes to remote models.
How to disable. Remove the codenib MCP server in the agent's settings and uninstall the package with pip uninstall codenib. Local indexes sit in the project and can be deleted along with the index folder.
FAQ
How do I install CodeNib for Claude Code?
Run python -m pip install "codenib[graph,mcp]", then codenib codegraph init with the path to a clean repository. The command builds the local search and graph and connects installed agents. After that ask the agent to use explore_context.
Does CodeNib need an API key or its own model?
No. The main CodeGraph path needs no model, key or GPU: your agent's model does the reasoning. The exception is the optional grep and Jev route in version 0.2.4, which works through OpenRouter.
Which languages does the graph support?
Graph backends exist for Python, Go, Rust, C and C++, C#, Java, Ruby, PHP, Kotlin, Scala, JavaScript and TypeScript. Swift and Lua get chunking and retrieval. The capability matrix by language is in the documentation.
How is CodeNib different from Serena?
Serena searches by symbol name or regular expression and gives one level of references per call. CodeNib answers a plain-language question and builds a call chain up to eight steps deep in one call. The README gives this comparison.
What does the dependency_subgraph tool do?
It builds a dependency subgraph from the chosen function: incoming and outgoing static references up to eight steps deep. Check the result by reading the source, because dynamic relations may be missing.
Discussion
No questions or reviews yet.
Related
MCP serversIndexes a codebase into a knowledge graph and gives the agent call chains, change blast radius and code search over MCP
Local code graph for agents: an MCP server returns entry points, symbol links and change impact in one call
A local semantic code index: the agent searches by meaning, then follows hits into definitions, callers and dependency paths
Source search ranks matches using code structure and gives agents relevant snippets without a separate search index.
A local codebase index for agents: hybrid search, dependency graphs and change impact analysis
An MCP server that builds a code graph of functions, classes, calls and routes and answers the agent's structural queries locally
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| CodeNib (this card) | yes | not required | 91 | |
| Serena | no | not required | 31,148 | searches by symbol name and gives one level of references per call through a live language server |
| CodeGraph | no | not required | 76,797 | a code graph built on syntax tree parsing, while CodeNib relies on SCIP and LSP indexers |
README in short
The README starts with a reranking experiment result and a two-command quickstart. It continues with arguments for the project, a comparison table against grep, Serena, Colby McHenry's CodeGraph and the public DeepWiki MCP, a language list and a table of links to benchmarks and reproductions. It ends with the beta status, citation and the Apache 2.0 licence.