Local MCP server that converts files into searchable AI context
stashbase, created by Liliu Z, converts personal files into a local knowledge source for AI agents by indexing documents and exposing context to compatible clients. The app builds a searchable local index and supplies context snippets for assistant prompts, with on-device text extraction and file-based wiki outputs. It targets developers, researchers, and AI power users who need private, agent-accessible project documentation and research libraries. Installation and development require a Node.js environment when building from source.
Creates linkable Markdown pages from diverse source material
The app automatically extracts text and assembles a navigable file-based wiki from your originals, producing ordinary Markdown files you keep. It supports multiple input types, letting users ingest research artifacts into a persistent archive. Supported formats include:
- PDF and DOCX
- Images (OCR-derived text)
- Audio recordings (transcribed text)
Those outputs make it easier for external agents to open and reference concrete documents rather than opaque blobs.
Search accuracy hinges on extraction quality and vector indexing
The tool employs vector-retrieval methods so queries match meaning instead of only exact words, allowing agents to find relevant passages when phrasing differs. That retrieval quality depends on the fidelity of OCR and transcription when working from images or recordings, and on how source material is chunked during indexing. Users should expect better matches from clean, well-structured documents than from noisy scans or poor audio.
Setup and workflow assume developer-level familiarity
Built with TypeScript and Electron, the project expects a Node.js environment and the pnpm workflow for users who compile from source. The app functions as a server compatible with MCP-aware clients such as Claude Desktop and Cursor, so adopting it requires connecting it to an existing agent workflow. Scoped access controls help map local folders into an agent's searchable scope, which suits technical users who manage deployment steps.
Local-first design preserves ownership and supports auditability
The app keeps generated files on the host machine as ordinary files that users own, and it offers folder-level scoping so agents only see explicitly allowed content. The project is open-source and developed as a human-directed experiment by the developer, and it has attracted attention on GitHub for its approach to persistent local memory for agents. That combination favors users who prioritize file ownership and traceability.
Best for hands-on developers and researchers who value local control
The app is a practical choice for technically minded users because it is an open-source, human-directed project with visible traction on GitHub. Expect to manage build and indexing details yourself and to monitor extraction quality for mission‑critical queries. For smoother results, prepare source documents with consistent headings and clean text so the generated wiki links and vector index produce more reliable context for agents.
