
AIRP-Core
by ghostxia
Independent, open-source LEGO-style Agent backend with self-orchestrated LLM loops and state persistence.
What it does
AIRP-Core is a high-performance, independent Agent backend designed as the "reasoning brain" of the AIRP ecosystem. Unlike simple MCP servers, Core manages its own LLM calls and executes complex, bounded agent loops on the server side. It specializes in role-playing (RP) and state-driven interactions, supporting SillyTavern V2 character cards and lorebooks while ensuring the agent remains within specific constraints (the "Bounded Agent Commandments").
Tools
plugin_kv_get/set: Get or set key-value data within a specific plugin namespace.plugin_jsonl_append/read: Perform O(1) append or paginated read of JSONL logs.plugin_blob_write/read: Write or read arbitrary binary/text blobs up to 4MiB.get_character/get_live_state: Retrieve character metadata and real-time state.seal_volume: Persist the current context into a numbered volume for long-term memory.create_scene/add_scene_character: Manage multi-character scenarios and configurations.
Installation
Run as a daemon to expose an OpenAI-compatible SSE gateway:
{
"mcpServers": {
"airp-core": {
"command": "airp-core",
"args": ["daemon", "--port", "8000"]
}
}
}
Supported hosts
Confirmed to work as a standalone backend; compatible with any OpenAI-compatible client including Claude Desktop via the provided SSE gateway.
Quick install
cargo run -- daemon --port 8000Information
- Pricing
- free
- Published
- 7/27/2026
- stars
- 0
Categories
Choose your AI client and follow the steps below.
Claude Desktop
Configure as an HTTP server pointing to the airp-core daemon port (default 8000).





