Memory HTTP

This MCP server provides an HTTP streaming memory-backed knowledge graph. It stores entities, relations, and observations in a memory-backed JSON format, supports streaming updates over HTTP, and exposes a straightforward MCP-compatible API for building, querying, and updating knowledge graphs in a self-contained HTTP service.

How to use

You run the HTTP streaming MCP server locally and connect your MCP client to it. Use it to create entities and relations, add observations, read the whole graph, search nodes, and monitor real-time updates via server-sent events. All interactions follow the MCP protocol and you manage sessions with a session header.

How to install

Prerequisites you need on your machine:

• Node.js installed (recommended version compatible with your tooling)
• pnpm installed globally (npm i -g pnpm)

Step by step, set up and run the server:

# 1) Navigate to the memory HTTP server directory
cd src/memory-http

# 2) Install dependencies
pnpm install

# 3) Build the project
pnpm build

# 4) Start the HTTP MCP server
pnpm start

Additional sections

Configuration and runtime environment can influence how you deploy and run the server. The server starts by default on port 3002 and exposes its MCP endpoint at /mcp. You can override the port and memory store path via environment variables when you start the service.

Environment variables you can use:

• PORT: Server port (default 3002)
• MEMORY_FILE_PATH: Path to the memory storage file (default ./memory.json)

Available tools

create_entities

Create new entities in the knowledge graph with specified names, types, and initial observations.

create_relations

Establish relationships between entities by specifying a from-entity, a to-entity, and a relation type.

add_observations

Add new observations to existing entities to enrich their context.

delete_entities

Remove entities and their related relations from the knowledge graph.

delete_observations

Remove specific observations from an entity.

delete_relations

Delete a specific relationship between two entities.

read_graph

Read and return the entire knowledge graph in its stored form.

search_nodes

Search for entities by a query string, matching on names or observations.

open_nodes

Retrieve entities by exact name match for quick access.