MCP Server

Litica exposes two tools over the Model Context Protocol for writing to and retrieving from the shared namespace. Any MCP-compatible agent can use them without code changes.

Overview

The Model Context Protocol is a standard for connecting external tools to AI models. The Litica MCP server implements this protocol to give any compatible agent a persistent, structured namespace for storing and retrieving context. Once configured, the agent calls these tools automatically when it determines they are relevant to the current task.

The server supports two transports: HTTP (for production and remote deployments) and stdio (for local Claude Desktop use). Both expose the same two tools. Each agent writes to its own namespace viaagent_id.

add_memory

Writes structured context to the agent's namespace. Litica decomposes the content into reference frames (goals, entities, emotional valence, temporal scope), embeds each domain independently with a local sentence-transformers model, and stores the result in PostgreSQL with pgvector. Provenance is recorded at write time.

parameters

content: str       # The experience to store (required)
agent_id: str      # Agent identifier, default "default"

Returns a confirmation string when the memory is stored.

search_memory

Retrieves context relevant to a natural language query using spreading activation. Litica generates a reference frame for the query, activates the namespace across all four domains in parallel, aggregates scores using domain-specific weights, and applies a salience multiplier that surfaces context that has proven useful before. Returns ranked results with provenance.

parameters

query: str         # Natural language query (required)
top_k: int         # Number of results, default 5
agent_id: str      # Agent identifier, default "default"

Returns a list of memory text strings ordered by relevance.

How retrieval works

When search_memory is called, Litica does the following:

Generates a reference frame for the query using Gemini
Queries each domain table (goals, entities, emotional valence, temporal scope) in parallel using cosine similarity against the stored embeddings
Aggregates scores across domains with weighted combination (goals 30%, entities 25%, emotional valence 20%, temporal scope 25%)
Applies a salience multiplier based on emotional importance and retrieval frequency
Returns the top-k results by final score

Context that is retrieved often accumulates salience and surfaces more readily in subsequent searches. This is the spreading activation model: relevance propagates through the namespace rather than being computed purely at query time.

Compatibility

The Litica MCP server works with any MCP-compatible host, including:

Claude Desktop (HTTP or stdio transport)
Claude Code CLI (HTTP transport via .mcp.json)
Cursor (HTTP transport)
Any custom agent built with the MCP SDK

Ready to connect? See the configuration guide.