Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@hermes-memory-mcp store user preference: concise answers in semantic tier"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A high-fidelity, multi-tier memory system designed for the SuperInstance Ecosystem. This Model Context Protocol (MCP) server provides a standardized interface for autonomous agents to persist, retrieve, and manage cognitive states across different levels of abstraction.
The core innovation of this system is its non-linear, multi-tier storage architecture. Rather than a flat "key-value" store, Hermes organizes intelligence into three distinct tiers:
Tier | Cognitive Level | Description | Use Case |
Semantic | The "Why" & "How" | High-level reasoning, design principles, and philosophical intent. | "Why was the sensor fusion logic implemented this way?" |
Structural | The "Who" & "What" | Entity maps, agent taxonomies, repo structures, and SOPs. | "What is the current hierarchy of the Claw fleet?" |
Technical | The "Where" & "How Much" | Absolute paths, API schemas, hardware constraints, and environmental facts. | "What is the absolute path to the sonar logs?" |
Related MCP server: Strata Memory MCP Server
Agents interacting with this server have two primary superpowers:
Purpose: Commit a new piece of intelligence to the substrate.
Parameters:
: ( | | ) — The ownership domain.
: ( | No files replaced | ) — The modification type.
: ( | | ) — The cognitive layer.
: The actual declarative fact or instruction.
: Metadata for high-speed retrieval.
Purpose: Perform targeted retrieval of intelligence.
Parameters:
: Natural language or keyword search.
: Filter by cognitive level to reduce noise.
: Filter by ownership.
To maximize the efficacy of this memory center, agents must adhere to these Communication Protocols:
Agents must avoid writing "intent" or "transient thought" to the memory substrate. Memory should only contain stabilized intelligence.
Bad Practice (Transient/Intent) | Good Practice (Declarative Fact) |
"I am going to try to find the log file." | "Target: Log directory identified at ." |
"I think the user wants concise answers." | "User preference: Concise, technical-first communication styles." |
"I will record the sensor data every 10m." | "SOP: Sensor telemetry is recorded at 10-minute intervals." |
When an agent enters a "Task Completion" state, it must perform a Memory Handover to ensure its successor doesn't start from zero.
Identify: What was learned? (e.g., a new API endpoint).
Tier: Which tier does it belong to? (e.g., ).
Commit: Call .
Install Dependencies: up to date, audited 18 packages in 949ms
3 packages are looking for funding
run npm fund for details
1 high severity vulnerability
To address all issues (including breaking changes), run: npm audit fix --force
Run npm audit for details.
2. Build:
hermes-memory-mcp@1.0.0 build tsc
Configure MCP Client: Add the following to your configuration (e.g. Claude Desktop):
This server cannot be deployed
Persistent long-term memory for AI agents: semantic search, knowledge graph, and task canvas.
Long-term memory for AI agents: semantic facts, episodic events, and procedural workflows
Universal memory runtime for AI agents — episodic, semantic, and procedural memory.
Long-term memory for AI agents: durable records, observable retrieval, governed context assembly.
