How do I install the MCP Vstash skill?

Run npx skillmds@latest add jr2804/mcp-vstash in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is MCP Vstash safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: makes network calls, reads secrets. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with MCP Vstash?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is MCP Vstash free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published MCP Vstash?

jr2804 (@jr2804) published this skill. Their other Agent Skills are listed on their SkillMD profile.

MCP Vstash by jr2804 — Agent Skill

Name: MCP Vstash
Rating: 4.2 (5 reviews)
Author: jr2804

MCP Vstash

MCP server integration for vstash document memory. Use when configuring Claude Desktop or other MCP-compatible AI assistants with persistent document memory, setting up vstash MCP tools for semantic search and Q&A, or integrating vstash with AI assistant workflows via Model Context Protocol.

jr2804 Updated Sep 10, 2026

File contents

vstash MCP Server

Enable AI assistants (Claude Desktop, Cursor, Copilot) to search and answer questions from your local document memory via MCP.

Setup

1. Install vstash

pip install vstash

2. Add to Claude Desktop config

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "vstash": {
      "command": "vstash-mcp"
    }
  }
}

pyenv users: Use full path to binary: "command": "/path/to/.pyenv/versions/3.x.x/bin/vstash-mcp"

3. Restart Claude Desktop

The vstash tools appear in Claude's tool list.

Available Tools

Tool	Description
`vstash_add(path)`	Ingest file, directory, or URL into memory
`vstash_ask(query, top_k)`	Semantic search + LLM answer with sources
`vstash_search(query, top_k)`	Hybrid search with context expansion and relevance signal
`vstash_list()`	List all ingested documents
`vstash_stats()`	Database statistics (docs, chunks, size)
`vstash_forget(source)`	Remove document from memory
`vstash_collections()`	List all collections
`vstash_export(...)`	Export chunks as JSONL for training data curation
`vstash_job(job_id)`	Check status of background directory ingestion

Search Response Fields

vstash_search returns:

Field	Description
`chunks`	Array of results with ±1 adjacent chunks expanded
`relevance`	Confidence tier: `"high"`, `"medium"`, `"low"`, `"none"`
`hint`	Human-readable relevance explanation
`best_distance`	Cosine distance of best vector match (lower = more relevant)

Relevance tiers:

Distance	Tier	Meaning
≤ 0.95	high	Results are relevant
0.95–0.98	medium	Results may be tangential
> 0.98	low	Results may not be relevant

Works from first search — no warm-up period needed.

API Key Configuration

MCP servers don't inherit shell environment variables. Configure inference in ~/.vstash/vstash.toml:

[cerebras]
api_key = "your-key-here"

Or use fully local Ollama (no API key):

[inference]
backend = "ollama"

[ollama]
host = "http://localhost:11434"
model = "llama3.2"

Troubleshooting

Issue	Solution
Tools don't appear	Run `which vstash-mcp` to verify PATH
"No module named vstash"	MCP server uses different Python — use full path in config
`ask` fails but search works	Check inference backend configured in `vstash.toml`

Reference

See references/mcp-reference.md for tool-specific options and advanced configuration.

---name: mcp-vstashdescription: MCP server integration for vstash document memory. Use when configuring Claude Desktop or other MCP-compatible AI assistants with persistent document memory, setting up vstash MCP tools for semantic search and Q&A, or integrating vstash with AI assistant workflows via Model Context Protocol.---# vstash MCP ServerEnable AI assistants (Claude Desktop, Cursor, Copilot) to search and answer questions from your local document memory via MCP.## Setup### 1. Install vstash```bashpip install vstash```### 2. Add to Claude Desktop configEdit `~/Library/Application Support/Claude/claude_desktop_config.json`:```json{ "mcpServers": { "vstash": { "command": "vstash-mcp" } }}```> **pyenv users:** Use full path to binary:> `"command": "/path/to/.pyenv/versions/3.x.x/bin/vstash-mcp"`### 3. Restart Claude DesktopThe vstash tools appear in Claude's tool list.## Available Tools| Tool | Description ||------|-------------|| `vstash_add(path)` | Ingest file, directory, or URL into memory || `vstash_ask(query, top_k)` | Semantic search + LLM answer with sources || `vstash_search(query, top_k)` | Hybrid search with context expansion and relevance signal || `vstash_list()` | List all ingested documents || `vstash_stats()` | Database statistics (docs, chunks, size) || `vstash_forget(source)` | Remove document from memory || `vstash_collections()` | List all collections || `vstash_export(...)` | Export chunks as JSONL for training data curation || `vstash_job(job_id)` | Check status of background directory ingestion |## Search Response Fields`vstash_search` returns:| Field | Description ||-------|-------------|| `chunks` | Array of results with ±1 adjacent chunks expanded || `relevance` | Confidence tier: `"high"`, `"medium"`, `"low"`, `"none"` || `hint` | Human-readable relevance explanation || `best_distance` | Cosine distance of best vector match (lower = more relevant) |**Relevance tiers:**| Distance | Tier | Meaning ||----------|------|---------|| ≤ 0.95 | high | Results are relevant || 0.95–0.98 | medium | Results may be tangential || > 0.98 | low | Results may not be relevant |Works from first search — no warm-up period needed.## API Key ConfigurationMCP servers don't inherit shell environment variables. Configure inference in `~/.vstash/vstash.toml`:```toml[cerebras]api_key = "your-key-here"```Or use fully local Ollama (no API key):```toml[inference]backend = "ollama"[ollama]host = "http://localhost:11434"model = "llama3.2"```## Troubleshooting| Issue | Solution ||-------|---------|| Tools don't appear | Run `which vstash-mcp` to verify PATH || "No module named vstash" | MCP server uses different Python — use full path in config || `ask` fails but search works | Check inference backend configured in `vstash.toml` |## ReferenceSee `references/mcp-reference.md` for tool-specific options and advanced configuration.

Frequently asked questions