Quick Start: MCP Server
Kronroe's MCP server gives Claude Desktop, Cursor, and any MCP-compatible AI assistant persistent, bi-temporal memory. It wraps the AgentMemory API over a stdio transport with LSP-style Content-Length framing -- no separate database server required. All memory is stored in a single local file.
Kronroe’s MCP server gives Claude Desktop, Cursor, and any MCP-compatible AI assistant persistent, bi-temporal memory. It wraps the AgentMemory API over a stdio transport with LSP-style Content-Length framing – no separate database server required. All memory is stored in a single local file.
Installation
Choose one of three installation methods:
Fastest route if you already use Node.js tooling.
npx kronroe-mcpThis delegates to the `kronroe-mcp` binary on your PATH.
Use this if your workflow already centers on Python.
pip install kronroe-mcp
kronroe-mcpSet `KRONROE_MCP_BIN` to point at a custom binary location if needed.
Use this if you want to build the server from source.
cargo install --path crates/mcp-serverClaude Desktop Configuration
Add the following to your Claude Desktop MCP configuration file (claude_desktop_config.json):
{
"mcpServers": {
"kronroe": {
"command": "kronroe-mcp",
"env": {
"KRONROE_MCP_DB_PATH": "~/.kronroe/memory.kronroe"
}
}
}
}
Cursor Configuration
Add Kronroe as an MCP server in Cursor’s settings:
{
"mcpServers": {
"kronroe": {
"command": "kronroe-mcp",
"env": {
"KRONROE_MCP_DB_PATH": "~/.kronroe/memory.kronroe"
}
}
}
}
Database Path Configuration
The server stores all memory in a single file. Configure the path with the KRONROE_MCP_DB_PATH environment variable:
export KRONROE_MCP_DB_PATH=/path/to/memory.kronroe
If unset, the server defaults to ./kronroe-mcp.kronroe in the current working directory.
Basic Usage
Once configured, the MCP server exposes 11 tools to your AI assistant. Here is a typical workflow:
Storing memory with remember
Tell your assistant something and it can store it:
User: "Alice works at Acme Corp as a senior engineer."
The assistant calls remember with:
{
"text": "Alice works at Acme Corp as a senior engineer."
}
The server parses the text, extracts facts, and stores them with full bi-temporal metadata.
Retrieving memory with recall
Later, ask your assistant a question and it searches memory:
User: "Where does Alice work?"
The assistant calls recall with:
{
"query": "where does Alice work",
"limit": 5
}
The server performs a full-text search across all stored facts and returns matching results ranked by relevance.
Structured assertions with assert_fact
For precise, structured facts:
{
"subject": "alice",
"predicate": "works_at",
"object": "Acme Corp",
"confidence": 0.95,
"source": "user_statement"
}
Checking what you know with facts_about
Retrieve all current facts about an entity:
{
"entity": "alice"
}
Available Tools
The server exposes 11 tools: remember, recall, recall_scored, assemble_context, facts_about, assert_fact, correct_fact, invalidate_fact, what_changed, memory_health, and recall_for_task. See the MCP Tools Reference for full parameter documentation.