#473473rd of 489 in Skills, plugins and prompts

stash

Persistent memory server for MCP agents, backed by Postgres and pgvector

Stars
772
License
Apache-2.0
Last commit
Jun 2026
Last release
May 2026
Language
Go
Stars, last 29 weeks+527
25 Apr 2026: 245 stars2 May 2026: 615 stars9 May 2026: 660 stars16 May 2026: 681 stars23 May 2026: 687 stars30 May 2026: 697 stars6 Jun 2026: 701 stars13 Jun 2026: 703 stars20 Jun 2026: 705 stars27 Jun 2026: 712 stars4 Jul 2026: 716 stars11 Jul 2026: 738 stars18 Jul 2026: 748 stars25 Jul 2026: 753 stars1 Aug 2026: 753 stars8 Aug 2026: 756 stars15 Aug 2026: 759 stars22 Aug 2026: 761 stars29 Aug 2026: 761 stars5 Sept 2026: 763 stars12 Sept 2026: 763 stars19 Sept 2026: 765 stars26 Sept 2026: 766 stars3 Oct 2026: 768 stars10 Oct 2026: 768 stars17 Oct 2026: 768 stars24 Oct 2026: 768 stars31 Oct 2026: 770 stars7 Nov 2026: 770 stars14 Nov 2026: 772 stars

245 on 25 Apr 2026772 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

Stash is a Go service that exposes an MCP server over SSE so agents can store and recall memories across sessions. A background 9-stage consolidation pipeline uses an embeddings model and a chat model to turn raw episodes into facts, relationships, causal links, patterns and contradictions. It runs through Docker Compose with Postgres and pgvector, and works with OpenAI-compatible APIs or local Ollama.

Who it is for: Developers adding long-term memory to MCP-capable coding agents or assistants

Strengths

  • One docker compose up starts Postgres with pgvector, migrations and the MCP server
  • Works with any MCP client over SSE; README gives Cursor, Claude Desktop, Windsurf and OpenCode configs
  • Consolidation stages process only data added since the previous run
  • Can run fully local with Ollama, or against any OpenAI-compatible base URL

Weaknesses

  • Requires both an embeddings model and a chat model for consolidation
  • Changing embedding model means setting STASH_VECTOR_DIM to match the new dimension
  • Hosted cloud version shares no code and has a different feature set
  • Transport is SSE only; no stdio mode is documented

What it needs

  • Docker + Compose
  • Needs Postgres with pgvector, OpenAI-compatible embeddings and chat API, or Ollama, Docker Compose
  • Models: OpenAI-compatible APIs, Ollama, Atlas Cloud
  • port 8080

What changed

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in Skills, plugins and prompts

See all 489
Also in Skills, plugins and prompts
RankProjectScore
1claude-memPersistent cross-session memory plugin for Claude Code and other coding agents118.1k starsApache-2.097 out of 100
2impeccableDesign skill with 24 slash commands for AI coding agents' frontend output Live demo ↗ (opens in a new tab)112.4k starsApache-2.097 out of 100
3gstackMarkdown slash-command skills that give Claude Code planning, review, QA and ship roles141.9k starsMIT96 out of 100
#4rufloAgent harness adding swarms, memory and hooks to Claude Code and Codex Live demo ↗ (opens in a new tab)82k starsMIT96 out of 100
#5eccAgents, skills, hooks and memory bundle for Claude Code and other harnesses305.7k starsMIT95 out of 100