FSKB MCP
FSKB MCP, bundled in LLM Tornado, indexes code directories and gives agents semantic search with local or hosted embeddings.
- 654 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The indexer reads source files and stores vectors; hosted embedding providers receive the selected code.
Install
- Works in
Manual install
pip install -r requirements.txt
pip install -e .Installs dependencies and FSKB from `src/LlmTornado.FsKb/LlmTornado.FsKb`. This is the subproject README’s first installation sequence after entering its directory; we did not run it.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The indexer reads source files and stores vectors; hosted embedding providers receive the selected code.
This is third-party code. Review the repository files before installing.
What it does
FSKB indexes repository files and gives an agent semantic search through stdio MCP. For example, an agent can find error handling without knowing a function’s name. The indexer respects Git ignore rules and skips heavy generated directories. ChromaDB stores the index locally. The program watches files and updates it after edits. It also offers a command-line and graphical interface. FSKB lives in a separate directory of the LLM Tornado repository and installs as a Python component.
Who it is for. For developers who want an agent to search their own repositories semantically.
Good fit when
- Find an implementation by purpose without its exact name.
- Include several directories in one search collection.
- Keep the index current after edits.
- Choose local embeddings for private source code.
Not a fit when
- You need a call graph with complete dependency coverage.
- Code must not reach the chosen embedding API.
- You expect a ready server without configuring models and directories.
Example request
Index the test directory with FSKB and locate this operation’s handling, returning source passages without editing code.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Python 3.10 or later and the FSKB subproject dependencies are required; they install separately from the main LLM Tornado library. Local search needs downloaded model weights; choosing a hosted API sends code to that provider using your key. Extension and size filters determine coverage: excluded files will not appear in search results. Indexes and logs remain in a separate directory after the server stops; manage their retention and deletion separately from the source files.
How to disable. Stop `fskb --mcp` and remove the client connection. Indexes and logs remain in the configured data directory.
FAQ
Which MCP transport is used?
stdio. The client starts installed `fskb` with `--mcp`.
How do I add a directory?
Use `fskb --add-root /path/to/repository` with the intended repository path.
Is the GUI required?
No. `fskb --no-gui` runs without it; MCP is another mode.
Where is data stored by default?
In `~/.fskb/data`, configurable in storage settings.
Can I change embedding providers?
Yes. Local models and hosted APIs through LiteLLM are supported through embedding configuration.
How mature is FSKB?
The version 0.1.0 manifest classifies it as alpha. Test indexing on a sample directory before connecting a working repository.
Discussion
No questions or reviews yet.
Related
MCP serversAn 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
A local codebase index for agents: hybrid search, dependency graphs and change impact analysis
An MCP server that indexes code with tree-sitter and returns individual functions and classes to the agent instead of whole files
A Python MCP server for BAF, a Ukrainian counterpart to 1C:Enterprise: configuration metadata, queries and full-text BSL code search
An MCP server that turns a large codebase into a semantic feature graph for precise navigation and meaning-based search
KotaDB indexes a repository locally and tells Claude Code what breaks when a file changes: dependencies and semantic search.
README in short
The example configuration separates storage, model provider and indexing constraints. GPU use depends on separately installed PyTorch with suitable CUDA support.