# dcostenco/prism-mcp [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/dcostenco/prism-mcp  
**GitHub Stars:** 156  
**npm Downloads (last month):** 5678  
**Views:** 8  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/dcostenco-prism-mcp

## Description
Zero-config persistent memory for AI agents with local SQLite. Mind Palace web dashboard, time travel (rewind/replay sessions), agent telepathy (cross-client memory sharing), code mode templates, morning briefings, and progressive context loading. 25 tools, 6 resources, 4 prompts.

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

```json
"mcpServers": {
  "prism-mcp": {
    "command": "npx",
    "args": ["-y","prism-mcp-server"]
  }
}
```

## Documentation

## What dcostenco/prism-mcp does

dcostenco/prism-mcp adds durable memory and local-first agent services to MCP-compatible clients. Session information survives restarts, allowing agents to retain handoff notes, recent work, open tasks, and other project context. Memory search can apply ACT-R ranking, combining relevance with recency, frequency, and associations; this behavior can be disabled in preferences or with `PRISM_ACTR_ENABLED=false`.

The server also provides a Mind Palace web dashboard for inspecting and managing the stored information. Its broader feature set includes cross-client memory sharing, session rewind and replay, code-mode templates, morning briefings, progressive context loading, offline tool routing, and a savings meter for locally served requests.

## How it works

The server runs locally and stores its persistent memory in SQLite. MCP clients access the functionality through the server's tools, resources, and prompts. The package exposes 25 tools, 6 resources, and 4 prompts.

Local routing can use installed Ollama models, with automatic selection across 2B, 4B, 9B, and 27B options based on available models, memory, context fit, and subscription entitlements. Eligible text coding tasks can also use a configured compatible oMLX server. The `local_savings` tool and the `prism savings` command report locally handled token volume, including local share and per-model detail, along with the assumptions behind the count.

## Setup and configuration

Install the `prism-mcp-server` npm package and connect it to supported clients with `prism connect`. The connection workflow can configure Claude Code, Claude Desktop, Cursor, Gemini CLI, and Codex while preserving unrelated settings. No account or API key is needed for the local server.

Use `prism dashboard` to open the web interface. Local AI settings are available under Account & Settings → Local AI, where users can select coding, review, and image backends and test saved model choices. Installing or updating the optional Prism-managed oMLX server is a separate dashboard operation and is supported on compatible Apple Silicon Macs.

## Tools and capabilities

- Persist and retrieve session memory across restarts.
- Share memory between connected AI clients.
- Search memories with optional ACT-R ranking.
- Review, rewind, and replay sessions.
- Route suitable work to local models without cloud inference.
- Record local usage and inspect token savings.
- Load context progressively instead of sending all stored material at once.
- Use code templates, morning briefings, and dashboard-based memory management.

## Limitations and notes

Local model behavior depends on the models installed and their available resources. A passing image diagnostic confirms execution, not visual accuracy. Compatible oMLX servers can handle controlled local PNG requests, while the saved Ollama image option currently does not execute. Image processing uses local screening, and model accuracy varies.

The npm package and an external oMLX server are updated independently. Updating the package or restarting an MCP client does not upgrade an external oMLX installation. A paid subscription adds cloud sync, higher model tiers, and team features through the Synalux portal; the local server itself does not require that subscription.

_Full upstream README: https://allmcps.com/mcp/dcostenco-prism-mcp/readme_