indra_db_mcp

Persistent memory for AI reasoning and decisions.

An MCP server that gives AI agents memory that persists across sessions. Built on indra_db — a git-like database for versioned thoughts.

The Problem

AI agents start fresh every session. Yesterday's insights evaporate. Decisions get re-made. Reasoning chains vanish.

indra_db_mcp changes that by giving agents:

• 🧠 Persistent memory — Record reasoning that survives session boundaries
• 🔍 Semantic search — Find past decisions by meaning, not keywords
• 🌿 Branching — Explore alternatives without losing the main thread
• 📜 History — See how understanding evolved over time

Quick Start

Install

No installation required — use bunx for automatic updates:

bunx -y indra_db_mcp@latest

Or install globally:

bun add -g indra_db_mcp

Configure Your Agent

Claude Code — Add to your project's CLAUDE.md:

@import node_modules/indra_db_mcp/INDRA_INSTRUCTIONS.md

Claude Desktop — Add to claude_desktop_config.json:

{
  "mcpServers": {
    "indra": {
      "command": "bunx",
      "args": ["-y", "indra_db_mcp@latest"]
    }
  }
}

OpenCode — Add to ~/.config/opencode/opencode.json:

{
  "mcp": {
    "indra": {
      "command": ["bunx", "-y", "indra_db_mcp@latest"],
      "type": "local"
    }
  },
  "instructions": ["~/.config/opencode/instructions/indra.md"]
}

Then copy INDRA_INSTRUCTIONS.md:

mkdir -p ~/.config/opencode/instructions
curl -o ~/.config/opencode/instructions/indra.md \
  https://raw.githubusercontent.com/moonstripe/indra_db_mcp/main/INDRA_INSTRUCTIONS.md

Generic MCP Client:

{
  "mcpServers": {
    "indra": {
      "command": "bunx",
      "args": ["-y", "indra_db_mcp@latest"],
      "env": {
        "INDRA_DB_PATH": "./.indra"
      }
    }
  }
}

Tools

Example Usage

An agent might use Indra like this:

User: Should I use PostgreSQL or MongoDB for my e-commerce app?

Agent thinking:
  → indra_search({ query: "database recommendations" })
  → Found: Previously recommended PostgreSQL for relational data with transactions
  
  → Making recommendation based on past reasoning + current context
  
  → indra_remember({ 
      content: "Recommended PostgreSQL for e-commerce app. User has relational product 
               catalog, needs transactions for orders. Consistent with past guidance.",
      id: "ecommerce-db-decision"
    })

Later:

User: Why did you recommend PostgreSQL?

Agent:
  → indra_search({ query: "ecommerce database" })
  → Found the reasoning from the previous session
  → Can explain the decision with full context

Branching for Exploration

Agent: Let me explore an alternative approach...

  → indra_experiment({ name: "try-nosql-approach" })
  
  [Explores MongoDB path, records reasoning]
  
  → indra_diff({ from: "main" })  // Compare with main reasoning
  
  → indra_branch({ action: "switch", name: "main" })  // Back to main

Environment Variables

How It Works

1. Content-addressed storage — Every entry is hashed. Identity comes from content.
2. Local embeddings — Uses sentence-transformers/all-MiniLM-L6-v2 for semantic search.
3. Git-like versioning — Commits create snapshots. Branches enable parallel exploration.
4. Single file — Everything in one .indra file. Easy to backup and share.

Architecture

┌─────────────────────────────────────────────────────────┐
│                    MCP Client (Claude)                   │
└─────────────────────────┬───────────────────────────────┘
                          │ MCP Protocol (stdio)
┌─────────────────────────▼───────────────────────────────┐
│                   indra_db_mcp (Bun)                     │
│  ┌─────────────┐  ┌──────────────┐  ┌───────────────┐  │
│  │   MCP SDK   │  │ IndraClient  │  │ Auto-sync     │  │
│  └─────────────┘  └──────┬───────┘  └───────────────┘  │
└──────────────────────────┼──────────────────────────────┘
                           │ CLI subprocess (JSON)
┌──────────────────────────▼──────────────────────────────┐
│                    indra CLI (Rust)                      │
│  ┌─────────────┐  ┌──────────────┐  ┌───────────────┐  │
│  │ Graph Store │  │  Embeddings  │  │  Git-like VCS │  │
│  └─────────────┘  └──────────────┘  └───────────────┘  │
└─────────────────────────────────────────────────────────┘

Visualize on IndraDB

Push to IndraDB for 3D visualization and analytics:

indra login
indra remote add origin username/my-memory
indra push origin

License

MIT

Related

• indra_db — The underlying Rust database
• IndraDB — Web platform for visualization
• MCP Specification — Model Context Protocol docs