harbor

# Harbor — Persistent Memory for AI Agents You now have access to Harbor, agent middleware that gives you persistent memory across sessions, credential isolation, and schema learning. ## Quick setup If Harbor is not installed, run the setup script: ```bash bash scripts/setup.sh ``` This installs the Harbor CLI and configures it as an MCP server for the current agent. If Harbor is already installed and configured as your MCP server, skip to **Using Harbor**. ## Using Harbor ### You have these MCP tools available | Tool | What it does | |------|-------------| | `harbor_remember` | Save context that persists across sessions and agents | | `harbor_recall` | Search and retrieve past context | | `harbor_http` | Auth-proxy HTTP — call any API without exposing credentials | | `harbor_learn_schema` | Teach Harbor which API fields matter — reduces noise permanently | | `harbor_get` | Query a connector (API) with auto-recall | | Connector tools | Each installed connector exposes tools like `coingecko_prices` | ### When you receive data from Harbor 1. **Check `meta.context`** — if present, another agent (or you in a past session) left analysis. Read it before reasoning. 2. **Check for `[Harbor:]` hints** — if you see one, call `harbor_learn_schema` immediately with 3-6 key fields. 3. **After analysis, save your findings** — call `harbor_remember` with a comprehensive summary. ### Auth-proxy HTTP (harbor_http) Call any API through Harbor's credential store — your agent never ...