Logic Server MCP server
{
"mcpServers": {
"nocturnus-logic-server": {
"command": "npx",
"args": [
"-y",
"nocturnusai-mcp@latest"
],
"env": {
"LLM_BASE_URL": "https://..."
}
}
}
}{
"mcpServers": {
"nocturnus-logic-server": {
"command": "npx",
"args": [
"-y",
"nocturnusai-mcp@latest"
],
"env": {
"LLM_BASE_URL": "https://..."
}
}
}
}{
"mcpServers": {
"nocturnus-logic-server": {
"command": "npx",
"args": [
"-y",
"nocturnusai-mcp@latest"
],
"env": {
"LLM_BASE_URL": "https://..."
}
}
}
}{
"servers": {
"nocturnus-logic-server": {
"command": "npx",
"args": [
"-y",
"nocturnusai-mcp@latest"
],
"env": {
"LLM_BASE_URL": "${input:llm_base_url}"
}
}
},
"inputs": [
{
"id": "llm_base_url",
"type": "promptString",
"description": "Llm base url",
"password": true
}
]
}{
"mcpServers": {
"nocturnus-logic-server": {
"command": "npx",
"args": [
"-y",
"nocturnusai-mcp@latest"
],
"env": {
"LLM_BASE_URL": "https://..."
}
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall npx -y nocturnusai-mcpversion 0.3.11
What it does
NocturnusAI is an MCP-compatible context engineering server that reduces token usage in multi-turn AI agent conversations by extracting structured facts, running deterministic backward-chaining inference, and returning only what changed since the last turn. It targets developers building AI agents or chatbots who want to reduce LLM input token costs without losing conversational accuracy.
- Reduce per-turn token counts in multi-turn agent conversations by sending only a delta of changed facts instead of full history replay.
- Use goal-driven inference to assemble only the facts relevant to a specific agent objective, such as determining SLA eligibility for a support ticket.
- Maintain a consistent logical state across conversation turns, automatically retracting derived conclusions when a source fact is retracted.
Tools this server exposes
| Tool | Does |
|---|---|
| context | Processes raw conversation turns and returns a salience-ranked working set of extracted facts as a briefing delta. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.