AutoMemBLOCK
MCP client for AutoMem — give Claude, Cursor, Codex, and other MCP tools durable graph+vector memory across conversations.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
One command. Infinite memory. Perfect recall across all your AI tools.
npx @verygoodplugins/mcp-automem setup
Your AI assistant now remembers everything. Forever. Across every conversation.
https://github.com/user-attachments/assets/fd79112b-5158-4320-a054-8c18ab1ea314
The guided installer — npx @verygoodplugins/mcp-automem install walks you through local, hosted, or existing-endpoint setup.
Works with Claude Desktop, Cursor IDE, Claude Code, GitHub Copilot (coding agent), ChatGPT, ElevenLabs, OpenAI Codex, OpenClaw, Hermes, Grok Build, Google Antigravity - any MCP-compatible AI platform.
The Problem We Solve
Every AI conversation starts from zero. Claude forgets your coding style. Cursor can't learn your patterns. Your assistant doesn't remember yesterday's decisions.
Until now.
AutoMem MCP connects your AI to persistent memory powered by [AutoMem](https://github.com/verygoodplugins/automem) - a graph-vector memory service.
What You Get
🧠 Persistent Memory Across Sessions
- AI remembers decisions, patterns, and context forever
- Works across all MCP platforms
043ae5892870OBSERVED · 2026-10-07Connect
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 mcp-automem -- npx -y @verygoodplugins/[email protected]
Exposed tools (28)
12 read · 14 write · 2 destructive. Blast radius: 2 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 |
|---|---|---|
AutoMem | read | Persistent AutoMem-backed graph memory for OpenClaw. |
bad-token-401-aborts | write | A 401 on the authed recall probe must abort before any write. |
claude-existing-headless | write | Sibling client: headless existing-target install for Claude Code in settings mode (the scriptable alternative to the recommended plugin). |
claude-plugin-autoinstall | read | Plugin mode with |
claude-plugin-fallback | read | Plugin mode with no |
cloud-instapods | read | hosted cloud → InstaPods: provider select → no resolve paste → setup-page plan |
cloud-other | read | hosted cloud → Other: provider select → paste URL + key up front |
cloud-railway | read | hosted cloud → Railway (guided): provider select → no paste → provision plan |
codex-existing-headless | write | Headless existing-target install for Codex against a healthy endpoint. |
delete_memory | destructive | Delete a memory by ID (\ |
dry-run-no-writes | write | Dry-run must produce a plan and change nothing. |
endpoint-500-aborts | write | A reachable-but-broken endpoint (500) must abort before any write. |
existing-claude-plugin | write | existing + Claude Code → plugin sub-prompt (manual step, no settings write) |
existing-claude-settings | read | existing + Claude Code → settings sub-prompt (writes settings.json) |
existing-cursor | read | existing endpoint, full interactive: target select → URL → key → agents (cursor) |
existing-grok | read | existing endpoint, full interactive: target -> URL -> key -> agents (grok, last in the list) |
existing-none | read | zero-agent guard: empty multiselect -> explicit confirm -> review says no agents |
grok-existing-headless | write | Headless existing-target install for Grok Build. Grok is deliberately absent from DEFAULT_AGENT_CLIENTS, so it is opt-in via --clients. |
idempotent-reinstall | write | Running the codex install twice must stay valid (no corruption on second pass). |
local | write | local docker: target → dir prompt → no agents (dry-run, no docker) |
malformed-health-200-html | read | Endpoint returns /health 200 with an HTML body (no JSON). Tests whether verify |
no-agent-install | write | --no-agent-install writes only .env, no client integration files. |
non-tty-no-yes-preview | write | Non-interactive without --yes/--dry-run must preview only and write nothing. |
recall_memory | read | Recall memories from AutoMem in one of three modes. The mode is selected by which params you pass. **Mode 1 — ID fetch:** pass \ |
store_memory | write | Store memory in one of two modes — single-memory (set top-level \ |
uninstall-after-install | destructive | Install codex via the real npx path, then |
update_memory | write | Update an existing memory |
website-bootstrap-install-sh | write | The real production entrypoint: website install.sh -> npx file:<tarball> install, headless. |
Trust audit
BLOCKgrade D · trust 68/100 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 | PASS |
| 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 (7 observation(s))
- Network
- declared (7 observation(s))
- Shell
- declared (6 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- found
Findings (25)
# Ports chosen to avoid the personal (:8001) and eval (:8031/:8041) stacks.
console.log(`✓ Capture B sidecar verified (answer cites ${PROVABLE_FACT.token})`);console.log(buildSummaryInstructions(endpoint, Boolean(apiKey)));
console.log(JSON.stringify(buildMcpConfigJson(endpoint, apiKey), null, 2));
apiKey = 'your-auto-mem-api-key'
delete_memory, uninstall-after-install
.commitlintrc.cjs
.git-blame-ignore-revs
.markdownlint.json
.prettierignore
.release-please-manifest.json
fileURLToPath(new URL('../../templates/claude-code', import.meta.url))fileURLToPath(new URL('../../templates/codex', import.meta.url))fileURLToPath(new URL('../../templates/copilot', import.meta.url))fileURLToPath(new URL('../../templates/COPILOT_INSTRUCTIONS_MEMORY_RULES.md', import.meta.url))fileURLToPath(new URL('../../package.json', import.meta.url))AUTOMEM_API_URL=http://127.0.0.1:8001
"AUTOMEM_API_URL": "http://127.0.0.1:8001"
- **AutoMem API URL** — e.g. `http://127.0.0.1:8001` or your Railway URL. Leave empty to use `AUTOMEM_API_URL` from your environment; falls back to `http://127.0.0.1:8001`.
"AUTOMEM_API_URL": "http://127.0.0.1:8001",
"AUTOMEM_API_URL": "http://127.0.0.1:8001",
@inquirer/prompts, @modelcontextprotocol/sdk, dotenv, node-fetch, smol-toml, yaml, @commitlint/cli, @commitlint/config-conventional
> relying on the bridge's `dotenv` to read `.env`. Avoids the secret, but is launch-cwd-fragile
curl -fsSL https://openclaw.ai/install.sh | bash
curl -fsSL https://automem.ai/install.sh | bash
Gates applied: no_behavioural_pass.
043ae5892870full audit observations/trust-audit/mcp-server/verygoodplugins__automem.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 |
|---|---|---|---|---|---|
| 2026-10-07 | 043ae5892870 | BLOCK | D | 68 | first audit |
Questions
What is the AutoMem MCP server?
MCP client for AutoMem — give Claude, Cursor, Codex, and other MCP tools durable graph+vector memory across conversations.
What tools does AutoMem expose?
28 in total: 12 read-only, 14 that write, and 2 that can delete or overwrite (delete_memory, uninstall-after-install). Every one is listed on this page with its risk.
Is AutoMem safe to connect to an agent?
No — not without reading the findings first. The audit graded it D (68/100) 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: 2 of its tools can destroy data, so scope the token you give it to what you actually need.
What credentials does AutoMem need?
It reads AUTOMEM_API_KEY, AUTOMEM_API_TOKEN, AUTOMEM_RECALL_TOKEN_BUDGET, MOCK_EXPECT_TOKEN and RAILWAY_API_TOKEN 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 AutoMem run?
It speaks sse, stdio and streamable-http, so it runs as a local process your client starts. It is published on npm as @verygoodplugins/mcp-automem at 0.16.0.
How current is this page?
The grade is for one exact copy of the source (043ae5892870), read on 2026-10-07. The repository is watched and re-audited when it changes.