TokenSave MCP
TokenSave is a local code MCP server that runs parsers in worker processes, isolating a grammar failure from the overall update.
- 704 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server reads source, stores an index and can edit files; setup adds client rules and hooks.
Install
- Works in
Manual install
cargo install tokensaveInstalls TokenSave through Cargo. Then connect an MCP client and index the selected project.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server reads source, stores an index and can edit files; setup adds client rules and hooks.
This is third-party code. Review the repository files before installing.
What it does
TokenSave is a local MCP server that builds a source-code graph in a local database and exposes results to agents through MCP. It can find symbols, trace calls and assemble task context from related fragments. Exploration starts with a focused query. After installation, users connect a client and initialize the project directory. The index tracks subsequent file changes, including writes made through targeted replacement and insertion tools. Separate Git worktrees support multiple agents working on a repository.
Who it is for. Engineers maintaining large codebases who need related source before changing a function or module.
Good fit when
- Before changing a function, a developer locates its callers.
- An agent receives source fragments related to a specific task.
- A team examines affected symbols while planning a refactor.
- After a targeted edit, a user refreshes the project graph.
Not a fit when
- A source index cannot be stored on disk.
- You require a perfect graph for every language construct.
- The client cannot read retrieved source fragments.
Example request
Use TokenSave to find this handler’s callers and show source before editing.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The index occupies .tokensave and contains source information, so restrict access as you would to the project itself. Local processing does not prevent the model from receiving retrieved fragments; your AI client controls that transfer. Language coverage depends on the build tier and parser, so verify important relationships against source. Setup can add agent configuration and Git hooks; inspect them before enabling automatic indexing of new projects.
How to disable. Run `tokensave uninstall` to remove integrations and Git hooks. Delete .tokensave after stopping the server.
FAQ
Does indexing need an API key?
Indexing itself needs no external key. Retrieved fragments still reach the MCP client, whose settings determine whether they are sent to a model.
How do I choose Codex?
Use `tokensave install --agent codex`.
How do I update the index manually?
Run `tokensave sync` in the project.
Can I reduce the build size?
Yes. The default Cargo build includes all supported languages. See lite and medium commands under [installation alternatives](https://github.com/aovestdipaperino/tokensave#installation) and choose coverage for your project.
How do I disable counter uploads?
Run `tokensave disable-upload-counter`. This disables counter uploads but leaves version checks and global-counter reads as network requests.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that indexes code with tree-sitter and returns individual functions and classes to the agent instead of whole files
An MCP server that builds a code graph once and serves the answer to the agent instead of re-reading files
A local MCP server with call graphs and symbol search: the agent sees code structure and what an edit affects in advance.
An MCP server and CLI for code research: search and understand your project and all of GitHub with fewer tokens
An MCP server that builds a code graph of functions, classes, calls and routes and answers the agent's structural queries locally
An MCP server that indexes a local repository with tree-sitter and gives the agent code search, file lookup and a structure summary of a file
README in short
Lite, medium and full builds differ in language coverage and binary size. Match the supported languages to the files in your project when choosing a build.