Fleet Memory MCP server
{
"mcpServers": {
"rcll-fleet-memory": {
"command": "npx",
"args": [
"-y",
"fleet-memory-mcp@latest"
],
"env": {
"HINDSIGHT_URL": "https://...",
"FLEET_URL": "https://..."
}
}
}
}{
"mcpServers": {
"rcll-fleet-memory": {
"command": "npx",
"args": [
"-y",
"fleet-memory-mcp@latest"
],
"env": {
"HINDSIGHT_URL": "https://...",
"FLEET_URL": "https://..."
}
}
}
}{
"mcpServers": {
"rcll-fleet-memory": {
"command": "npx",
"args": [
"-y",
"fleet-memory-mcp@latest"
],
"env": {
"HINDSIGHT_URL": "https://...",
"FLEET_URL": "https://..."
}
}
}
}{
"servers": {
"rcll-fleet-memory": {
"command": "npx",
"args": [
"-y",
"fleet-memory-mcp@latest"
],
"env": {
"HINDSIGHT_URL": "${input:hindsight_url}",
"FLEET_URL": "${input:fleet_url}"
}
}
},
"inputs": [
{
"id": "hindsight_url",
"type": "promptString",
"description": "Hindsight url",
"password": true
},
{
"id": "fleet_url",
"type": "promptString",
"description": "Fleet url",
"password": true
}
]
}{
"mcpServers": {
"rcll-fleet-memory": {
"command": "npx",
"args": [
"-y",
"fleet-memory-mcp@latest"
],
"env": {
"HINDSIGHT_URL": "https://...",
"FLEET_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 fleet-memory-mcpversion 0.1.0
What it does
RCLL is a self-hosted shared memory store for teams of AI agents, built as a fork of Hindsight. It adds topic scoping (rooms), knowledge typing (halls), priority tiers (L0–L3 layers), compressed summaries (closets), and cross-agent bridges (tunnels) over a hybrid vector-and-lexical search backend. The read path never calls a language model, so recalls cost CPU time but no LLM tokens.
- Store and retrieve scoped memories across multiple agents in a fleet, filtering by topic room and knowledge type to avoid irrelevant results.
- Share an insight discovered by one agent with another agent via a cross-bank tunnel without duplicating the underlying data.
- Compress accumulated L3 archive memories into closet summaries with source pointers to reduce context consumption during recall.
Tools this server exposes
| Tool | Does |
|---|---|
| memory_retain | Save a memory to the store with automatic or explicit room and hall classification. |
| memory_recall | Perform scoped semantic search filtered by room, hall, and layer depth; never calls a language model. |
| memory_reflect | Synthesize facts and find patterns across stored memories, returning results with citations via an agentic LLM loop. |
| memory_compress | Generate closet summaries from accumulated facts within a room and hall. |
| memory_bridge | Create a cross-bank tunnel to share memories between two agent banks. |
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.