What is Harbor?
Harbor is open-source agent context infrastructure that provides shared memory, credential isolation, and schema learning for AI agents. It connects via the Model Context Protocol (MCP) and works with Claude Code, Gemini CLI, Cursor, Codex, Minimax, and any MCP-compatible client. According to McKinsey, AI agents will handle 75% of enterprise workflows by 2028. Harbor ensures these agents share context across sessions and models — what one agent learns, every agent knows. Agent context is the missing infrastructure layer between agents and their tools. Harbor normalizes every API response into a consistent data[] + meta{} + errors[] envelope, learns which fields matter (47 fields in, 3 fields out), and controls what each agent sees with field-level access policies. Apache 2.0 licensed. Install in one command: curl -fsSL https://harbor.oseaitic.com/install | bash
Harbor provides high-level context management for AI agents — surfacing only relevant context from conversation history instead of flooding the context window. Think of it as context engineering infrastructure: the layer that decides what each agent needs to know, so every tool call carries the right context at the right time.
Your agents remember._Agent context infrastructure that sits between your agents and their tools.
What one agent learns, every agent knows.
harbor_remember({
connector: "coingecko",
note: "BTC dominance rising to 58%. SOL underperforming vs ETH.",
author: "Claude Code"
})
> Saved. This context will surface automatically in future sessions.{
"data": [{ "id": "bitcoin", "price": 67234, "change_24h": 2.34 }],
"meta": {
"source": "coingecko",
"context": {
"summary": "BTC dominance rising to 58%. SOL underperforming.",
"author": "Claude Code",
"age": "33 minutes ago"
}
}
}Gemini picks up where Claude left off. Zero re-prompting.
How It Works
One layer between your agents and the data.
No SDK, no code changes. Harbor handles context management between your agents and their tools. MCP-native, but any function-calling agent works.
Your agent won't make the same mistake twice
First debug session: agent discovers an API rate-limit quirk. Next week, different agent, different model — it already knows. Harbor injects past discoveries automatically. No code change, no manual recall.
Your agents never touch API keys
Harbor holds credentials. Agents hold context. Keys live in Harbor's credential store — never in config files, never in prompts. Agents connect via MCP and get data without seeing secrets.
Gets smarter the longer you use it
Your agent calls an API that returns 47 fields but uses 3. After a few calls, Harbor learns which 3 matter — and only sends those. Every future call, by any agent, is automatically curated.
Get Started
Choose how you run Harbor.
Start local with the CLI. Scale with Cloud when you need sync, encryption, and team access. Same protocol, same tools.
Everything runs on your machine. Open source, no account needed.
curl -fsSL https://harbor.oseaitic.com/install | bash
claude plugin marketplace add oSEAItic/harbor
Also works with any agent that supports Agent Skills — SKILL.md is auto-indexed.
Set up Harbor for this project — instructions at github.com/oSEAItic/harbor/blob/main/AGENTS.md
- —Normalize + Curate + Govern
- —Local memory & schemas
- —Any MCP-compatible client
- —Apache 2.0 licensed
Everything in CLI, plus cloud infrastructure for teams and production.
harbor login
- +Cross-device memory sync
- +End-to-end encryption
- +Team shared context
- +Dashboard + audit logs
- +Schema federation
Log in once — memories, schemas, and credentials sync automatically. No vendor lock-in — export anytime.
FAQ
Frequently asked questions.
Harbor is open-source agent context infrastructure for context engineering at scale — shared memory, credential isolation, and schema learning for AI agents. It connects via MCP and works with Claude Code, Gemini CLI, Cursor, Codex, and any MCP-compatible client.
Harbor holds API keys in its credential store. Agents connect to Harbor via MCP and never see raw API keys. This prevents credential leakage through prompts or plaintext config files. For cloud sync, credentials are encrypted client-side with AES-256-GCM.
When an agent calls harbor_remember, the note is stored and attached as meta.context to every future call to that connector. Any agent — regardless of model or session — receives this relevant context automatically, including conversation history and past discoveries. What Claude learns, Gemini knows.
Your agent calls an API that returns 47 fields, but only uses 3. After a few calls, Harbor learns which fields matter and only sends those — keeping your agent's context window focused on what matters. Four density layers: raw, normalized, compact, summary. Every future call by any agent is automatically curated.
Yes. Harbor connects via MCP, so any agent that supports the Model Context Protocol can use it instantly. For agents that don't support MCP natively, Harbor also exposes a standard tool-call interface.