Current source · verified path

Quickstart

Build the current source, point one MCP client at its server, and confirm the 13 tools before writing your first memory.

Build the current source

The repository is the current distribution path. Clone it, build it, then seed the six local vaults.

terminal

git clone https://github.com/frankxai/Starlight-Intelligence-System
cd Starlight-Intelligence-System
npm install
npm run build
node dist/cli.js init --vaults

Distribution note: as verified on July 24, 2026, npm serves v6.0.1 while this repository is v8.3.0. Use the source path above for the current system.

Configure your tool

Every adapter speaks the same MCP protocol. Pick yours and paste.

Claude Codeproject config

Keep the server declaration with the project that uses it.

memory:: CLAUDE.md
config:: .mcp.json

mcp config

{
  "mcpServers": {
    "starlight-sis": {
      "command": "node",
      "args": ["/path/to/Starlight-Intelligence-System/dist/mcp-server.js"]
    }
  }
}

Cursoruser config

Use one absolute server path so every workspace reaches the same vault.

memory:: .cursorrules
config:: ~/.cursor/mcp.json

mcp config

{
  "mcpServers": {
    "starlight-sis": {
      "command": "node",
      "args": ["/path/to/Starlight-Intelligence-System/dist/mcp-server.js"]
    }
  }
}

CodexTOML config

memory:: AGENTS.md
config:: ~/.codex/config.toml

mcp config

[mcp_servers.starlight_sis]
command = "node"
args = ["/path/to/Starlight-Intelligence-System/dist/mcp-server.js"]

Gemini CLIuser config

memory:: GEMINI.md
config:: ~/.gemini/settings.json

mcp config

{
  "mcpServers": {
    "starlight-sis": {
      "command": "node",
      "args": ["/path/to/Starlight-Intelligence-System/dist/mcp-server.js"]
    }
  }
}

OpenCodeuser config

Keep the server local while models change above the memory layer.

memory:: AGENTS.md
config:: ~/.config/opencode/opencode.json

mcp config

{
  "mcp": {
    "starlight-sis": {
      "type": "local",
      "command": [
        "node",
        "/path/to/Starlight-Intelligence-System/dist/mcp-server.js"
      ]
    }
  }
}

Verify it works

Ask the server for its tool inventory over the same JSON-RPC boundary your client uses.

terminal

printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' \
  | node /path/to/Starlight-Intelligence-System/dist/mcp-server.js

# result.tools contains 13 sis_* tool definitions

Add your first entry

In any configured tool, just tell the agent to remember something. Under the hood it calls sis_append_entry.

terminal

// in Claude Code, just ask:
"Remember that we chose Next.js 16 for the App Router streaming work."

// or call the tool directly:
sis_append_entry(
  vault: "technical",
  content: "Chose Next.js 16 for App Router streaming",
  confidence: "high"
)

That entry is now a plain JSONL line in ~/.starlight/vaults/technical.jsonl. Every tool with the MCP configured can read it. Your memory compounds across every session.

See the architecture →Explore live vaults →

// in Claude Code, just ask: "Remember that we chose Next.js 16 for the App Router streaming work." // or call the tool directly: sis_append_entry( vault: "technical", content: "Chose Next.js 16 for App Router streaming", confidence: "high" )