# Verified Memory [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/python1/data-lore-hive  
**GitHub Stars:** 0  
**npm Downloads (last month):** 179  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/verified-memory

## Description
Offline read-only search over a checksum-verified, optionally signed memory pack, with exact quotes

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

```json
"mcpServers": {
  "verified-memory": {
    "command": "npx",
    "args": ["-y","@python1/verified-memory-mcp"]
  }
}
```

## Documentation

## What Verified Memory does

Verified Memory MCP server exposes local memory lookup through MCP. Its source is a memory pack whose contents are checksum-verified and may also carry a signature. Searches are read-only and return exact quotes, making the result traceable to stored text rather than a newly generated paraphrase.

The project belongs in the 🧠 Knowledge & Memory category for use cases where an agent needs to consult a local evidence store. It is most suitable when the application can work from an existing pack and should avoid modifying or fabricating memory during a lookup.

The underlying repository is a research prototype focused on evidence-gated answers from local documents. Its broader pipeline can ingest documents, preserve immutable snapshots, and record model calls and verdicts in an append-only SQLite event log. The MCP-facing description is narrower: search the verified memory pack and expose exact quotations.

## How it works

A memory pack is checked through its stored checksums before content is used. When signing is enabled, the pack can also be validated against its signature. A query searches the local pack without requiring a network connection, and the returned evidence is the matching source text rather than an unconstrained answer.

The repository's evidence model is fail-closed. Its document-answering pipeline withholds results when a quote cannot be found verbatim, when a source has changed since its snapshot, or when independent checks disagree. The documented query outcomes include supported answers, confirmed absence, review-required cases, unsupported output, and stale sources. These broader behaviors provide context for the Verified Memory MCP server, but the available material does not specify individual MCP tool names or parameters.

## Setup and configuration

The project requires Python 3.10 or newer for its local command-line components and uses only the Python standard library. The full offline test suite can run without a model or network. Optional model mode uses Ollama on loopback at `127.0.0.1:11434`; a model must already be available, because the code does not download one.

Claude Desktop has a downloadable `.mcpb` add-on. The README identifies version `0.3.0`, provides a SHA-256 checksum, and says the add-on was tested manually on macOS and Windows. Installation can be performed by opening the file with Claude or dragging it into Claude Desktop's Extensions settings. Claude Desktop displays its normal warning for add-ons not verified by Anthropic.

## Tools and capabilities

- Search a local memory pack offline.
- Keep access read-only.
- Verify pack content with checksums.
- Optionally validate a signed pack.
- Return exact source quotes.
- Support local agent use through MCP.

## Limitations and notes

Verified Memory MCP server is a prototype, not a reliability benchmark. The README describes small, fixed-seed regression measurements on one model family and advises reviewing the project's stated limits before relying on the results.

The provided material does not document Cursor, Windsurf, Cline, or other MCP clients, nor does it list MCP operation names, schemas, configurable environment variables, or a separate package license. The repository also contains broader replication, recovery, signed-release, and Telegram-export memory work, but those are distinct components and should not be assumed to be MCP capabilities.

A signed pack is optional, so deployments should decide whether checksum verification alone meets their evidence requirements. Model mode is also optional; basic offline tests and local pack handling do not require Ollama.

Verified Memory MCP server fits developers who need local, quote-backed retrieval and can accept a read-only, prototype-stage component. It is less suitable when the agent must write memories, rely on cloud-hosted data, or use a client whose support is not documented.

_Full upstream README: https://allmcps.com/mcp/verified-memory/readme_