# vectorize-io/hindsight [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/vectorize-io/hindsight  
**GitHub Stars:** 54464  
**npm Downloads (last month):** 0  
**Views:** 15  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/vectorize-io-hindsight

## Description
Hindsight: Agent Memory That Works Like Human Memory - Built for AI Agents to manage Long Term Memory

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "hindsight": {
    "command": "npx",
    "args": ["-y","@vectorize-io/hindsight-coding-agents"],
    "env": {
      "HINDSIGHT_API_LLM_API_KEY": "",
      "HINDSIGHT_API_LLM_PROVIDER": "",
      "HINDSIGHT_DB_PASSWORD": ""
    }
  }
}
```

**Requires environment variables:** `HINDSIGHT_API_LLM_API_KEY`, `HINDSIGHT_API_LLM_PROVIDER`, `HINDSIGHT_DB_PASSWORD` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What vectorize-io/hindsight MCP server does

vectorize-io/hindsight exposes Hindsight’s long-term memory capabilities to AI agents through MCP. Hindsight is intended for agents that need to accumulate knowledge and use it in later interactions, rather than only retrieve recent conversation history.

The memory system organizes information into memory banks. An agent can add observations, search previously retained information, or ask Hindsight to generate a response based on what it knows. This makes the server relevant to assistants, coding agents, and other applications that need persistent context across sessions.

## How it works

Hindsight centers on three operations:

- Retain stores information in a selected memory bank.
- Recall searches a bank for memories related to a query.
- Reflect generates a response using the bank’s accumulated knowledge and the query’s context.

For example, an agent can retain a fact such as a person’s employer, recall what that person does later, and use reflect to produce a more complete answer. Memory is kept in Hindsight rather than limited to the current agent conversation.

The underlying Hindsight service can run with an embedded database, Docker, a separately managed PostgreSQL database, Kubernetes, or Hindsight Cloud. Clients can also use the HTTP API directly. Hindsight supports Python, Node.js/TypeScript, Go, and command-line clients in addition to MCP-based access.

## Setup and configuration

Self-hosted deployments can be started with Docker, installed with `pip install hindsight-api`, or deployed with the published Helm chart. The quick-start configuration uses an LLM provider and API key, such as `HINDSIGHT_API_LLM_API_KEY` with OpenAI. The service exposes an API on port 8888 in the Docker example and a UI on port 9999.

Hindsight supports hosted providers, local runtimes, OpenAI-compatible endpoints, and gateway options. Provider configuration is selected through `HINDSIGHT_API_LLM_PROVIDER`. Hindsight Cloud provides a hosted endpoint at `https://api.hindsight.vectorize.io` for deployments that do not run the service locally.

MCP clients named for the project include Claude Code and Cursor. The repository also describes support for Windows, Linux, macOS, air-gapped setups, and multiple deployment models.

## Tools and capabilities

- Retain facts, events, and other information in a memory bank.
- Recall relevant stored memories with a natural-language query.
- Reflect over memory to generate a disposition-aware response.
- Maintain persistent agent knowledge across conversations.
- Use hosted or self-managed Hindsight infrastructure.
- Connect through Hindsight’s supported API clients and service endpoints.

## Limitations and notes

Hindsight requires an LLM provider configuration for its memory service, unless using a supported subscription-based provider or local model setup. Storage and deployment choices affect how the service is operated: self-hosted installations require infrastructure management, while Hindsight Cloud uses a hosted endpoint.

The system is designed around Hindsight memory banks and its retain, recall, and reflect workflow. Agents that only need short-lived conversational context may not need a persistent memory service.

_Full upstream README: https://allmcps.com/mcp/vectorize-io-hindsight/readme_