gmem

@yksanjo

Visit Server

12 days ago

Persistent typed memory MCP server for Solana programs and EVM contracts. Agents remember Programs, PDAs, Contracts, Decisions, Findings across sessions.

Overview Tools Comments

Overview

What gmem solves

AI coding agents (Claude Code, Cursor, Cline, Continue.dev) lose all project context between sessions. Every morning you re-paste program addresses, re-explain why Pyth vs Chainlink, re-pin Anchor versions, re-walk the agent through last week's audit findings.

gmem fixes this with typed project memory — first-class entities for the things that matter on-chain.

Entities

Program — Solana programs with IDL hash, Anchor version, deploy commit
Contract — EVM contracts (Base, Optimism, Polygon, Arbitrum, Ethereum) with reorder-invariant ABI hash
PDA — seeds, program reference, last-seen account state
Account — wallets and on-chain accounts touched by the project
Decision — architectural choices with rationale, alternatives, commit ref, made-by
Finding — audit/review findings with severity, status, source reference
Integration — external services (oracles, indexers, RPC) with auth context

Features

Auto-ingest Anchor workspaces via gmem.ingest_anchor
Auto-ingest Hardhat workspaces via gmem.ingest_hardhat (v1.1+)
SQLite + FTS5 BM25 ranking for semantic recall across history
Append-only versioning — history preserved, never overwritten
One DB per project — no shared state, no cross-project leakage
7 MCP tools exposed for read/write/query/ingest workflows
MIT-licensed, single binary install via npm

Install

Add to your Claude Desktop / Cursor / Cline config:

{
  "mcpServers": {
    "gmem": {
      "command": "npx",
      "args": ["-y", "@yksanjo/gmem"]
    }
  }
}

Or one-line global:

npm install -g @yksanjo/gmem

Server Config

{
  "mcpServers": {
    "gmem": {
      "command": "npx",
      "args": [
        "-y",
        "@yksanjo/gmem"
      ]
    }
  }
}

Build with ShipAny.