M

memory-cloud

kagura-ai/kagura-memory · v0.173.2 · Development & Workflow

by kagura-aiScan grade: GoodAwesome AI Plugins

MCP server: self-hosted persistent memory for AI agents, with remember and recall tools that keep project facts searchable across sessions

Protect this install with HOL Guard

Guard checks the skills and MCP servers a plugin adds on your machine and asks you before a risky action runs. The scores above are free, and no plan changes them.

Is memory-cloud a fit?

From the plugin's manifest and HOL's scan. Anything the plugin does not declare is shown as not declared.

Works with

  • Claude Code
  • MCP
  • 1 skill
  • Runs hooks on agent events

Before you install

Sign-in
Not declared
Settings it asks for
Kagura Memory MCP URL, Kagura API key (secret), Guardrail context id, Strongest hook action
Declared access
Not declared
License
MIT

Check before relying on it

  • Some security checks did not finish: Cisco skill scan. See the evidence
  • 2 high or critical findings from the scan. See the evidence
  • The publisher has not verified ownership on HOL.

What is memory-cloud?

memory-cloud is a published development & workflow plugin for AI coding agents that works with Claude Code and MCP, developed by kagura-ai and distributed through the HOL AI plugin registry. MCP server: self-hosted persistent memory for AI agents, with remember and recall tools that keep project facts searchable across sessions.

Canonical slug
kagura-ai/kagura-memory
Version
v0.173.2 · updated Nov 20, 2026

Registry snapshot

Support the project
GitHub Sponsors
Owner verification
Not verified
Marketplace source
.claude-plugin/marketplace.json
Scanner
HOL plugin scanner
Security result
Needs review
Digest verified
Yes
1 bundled skillRead, copy or download each SKILL.mdOpen skills

Trust and security

Trust & reputation

35

HOL Trust Score

Per-metric points (0–100 each) combined via a weighted average into the overall score.

Installability
100/100
Maintenance
65/100
MCP server setup
100/100
Plugin Security
83/100
Provenance
70/100
Publisher Quality
75/100

Provenance

Plugin root
.
Source repo
https://github.com/kagura-ai/memory-cloud
Source commit
b5c961f84411…
Publisher verified
No
Owner verified
Not owner verified
Agent Plugins 1.0
Not detected

Continuous scanner CI not detected

This plugin remains listed. Its overall trust score is reduced by 10% because security checks are not maintained in the source repository's CI.

Optional: maintain the scanner in the source repository's CI to receive the full trust score. Listing does not require that change.

README badge not detected

Plugin owners can add the HOL badge to the repository README from Guard Plugins. Distribution, not security evidence. It adds a 2% engagement adjustment to the overall trust score and never changes the security result.

Needs review
Security result
83
Security score
2
High findings
Provider
codex-plugin-scanner
Grade
B · Good
Version
3.24.2
codex:hol-plugin-scanner-gXyHus / cisco-skill-scanner: failedcodex:hol-plugin-scanner-gXyHus / cisco-mcp-scanner: skipped

[claude:.] Claude Plugin

18/18

5 checks

[claude:.] Security

9/16

6 checks

[claude:.] Operational Security

N/A

5 checks · not applicable

[claude:.] Code Quality

10/10

2 checks

[codex:.] Manifest Validation

25/31

9 checks

[codex:.] Security

9/16

9 checks

[codex:.] Operational Security

N/A

5 checks · not applicable

[codex:.] Best Practices

9/12

5 checks

[codex:.] Marketplace

15/15

3 checks

[codex:.] Skill Security

5/5

4 checks

[codex:.] Code Quality

10/10

2 checks

Published advisories

  • GHSA-3259-mmfp-p5w8GitHub security advisoriesPublished

    A workspace-scoped API key could reach memories, files, tags and contexts of another workspace its owner belongs to

  • GHSA-5964-g5w8-c2mfGitHub security advisoriesPublished

    An unlinked Google or GitHub identity can still sign in to the account it created

  • GHSA-667v-2xgx-92vmGitHub security advisoriesPublished

    Sleep report tools skip workspace and context authorization: removed or demoted members can read and roll back past Sleep runs

  • GHSA-7r9g-h22f-wh4jGitHub security advisoriesPublished

    Resource tokens (X-Resource-API-Key) written to the Caddy access log

  • GHSA-cr23-7w9g-fxf7GitHub security advisoriesPublished

    Context allow-lists were not applied to memory-id access and to suspended or demoted members

  • GHSA-fqr3-fc3x-qf9cGitHub security advisoriesPublished

    Provider-link OAuth callback is not bound to the browser session that started it

  • GHSA-gw2r-4ch3-jxghGitHub security advisoriesPublished

    Workspace invitation tokens written to access, error and usage logs

  • GHSA-m3h8-5qx6-68g5GitHub security advisoriesPublished

    TOTP second factor has no attempt limit; a correct password resets the failure counter

  • GHSA-m833-q262-qfc7GitHub security advisoriesPublished

    A stolen session can make its own OAuth provider primary after the link window and take over the account through email sync

  • GHSA-mwhm-837r-7m8wGitHub security advisoriesPublished

    REST recall returns a linked account's private-context memories without checking the caller's access to the context

  • GHSA-whf2-8r8q-72gjGitHub security advisoriesPublished

    Workspace transferred to a member during their erasure can be hard-deleted with other members' data

  • GHSA-wjgx-h2h8-75hgGitHub security advisoriesPublished

    rollback_session can restore a memory type outside the agent binding's type filter

  • GHSA-wvpq-9v37-mw4wGitHub security advisoriesPublished

    Workspace viewers could create, modify and delete memories and graph edges in shared contexts

GitHub security advisories are those the source repository has published. Checked .

Findings

highsecurityHARDCODED_SECRETReported 2 times

Potential secret material was detected in claude-skills/setup.md.

Remove the secret from source control and load it securely at runtime.

infomanifest-validationPLUGIN_JSON_INTERFACE_SHORTDESCRIPTION

The interface object is missing a valid "shortDescription" value.

Add a valid "shortDescription" field to the interface metadata.

infomanifest-validationPLUGIN_JSON_INTERFACE_LONGDESCRIPTION

The interface object is missing a valid "longDescription" value.

Add a valid "longDescription" field to the interface metadata.

infomanifest-validationPLUGIN_JSON_INTERFACE_CATEGORY

The interface object is missing a valid "category" value.

Add a valid "category" field to the interface metadata.

infomanifest-validationPLUGIN_JSON_INTERFACE_CAPABILITIES

The interface object is missing a valid "capabilities" value.

Add a valid "capabilities" field to the interface metadata.

infomanifest-validationPLUGIN_JSON_INTERFACE_ASSET_WEBSITEURL

The interface field "websiteURL" must use HTTPS or point to a safe in-repo asset.

Update "websiteURL" to use HTTPS or an existing relative asset path.

infomanifest-validationPLUGIN_JSON_INTERFACE_ASSET_PRIVACYPOLICYURL

The interface field "privacyPolicyURL" must use HTTPS or point to a safe in-repo asset.

Update "privacyPolicyURL" to use HTTPS or an existing relative asset path.

infomanifest-validationPLUGIN_JSON_INTERFACE_ASSET_TERMSOFSERVICEURL

The interface field "termsOfServiceURL" must use HTTPS or point to a safe in-repo asset.

Update "termsOfServiceURL" to use HTTPS or an existing relative asset path.

infomanifest-validationPLUGIN_JSON_INTERFACE_ASSET_COMPOSERICON

The interface field "composerIcon" must use HTTPS or point to a safe in-repo asset.

Update "composerIcon" to use HTTPS or an existing relative asset path.

infomanifest-validationPLUGIN_JSON_INTERFACE_ASSET_LOGO

The interface field "logo" must use HTTPS or point to a safe in-repo asset.

Update "logo" to use HTTPS or an existing relative asset path.

infomanifest-validationPLUGIN_JSON_INTERFACE_ASSET_SCREENSHOTS

The interface field "screenshots" must use HTTPS or point to a safe in-repo asset.

Update "screenshots" to use HTTPS or an existing relative asset path.

infobest-practicesCODEXIGNORE_MISSING

A .codexignore file helps prevent accidental inclusion of local artifacts and secrets.

Add a .codexignore file with generated assets, local state, and secret paths.

Install guidance

Install in Claude Code

Install through the Claude Code plugin marketplace.

Claude Code plugin docs

Optional: install HOL Guard for Claude Code first, so it can check this plugin's tools on your machine before Claude Code uses them.

  1. 1

    Add the marketplace

    Run this inside a Claude Code session. It registers the marketplace manifest shipped at the repository root. If the repository ships none, add the generated entry from the Advanced section below first.

    claude code
  2. 2

    Install the plugin

    Installs the kagura-memory entry from the kagura-memory-cloud marketplace.

    claude code
  3. 3

    Scripted alternative

    Non-interactive equivalent for scripts and CI pipelines. Add --scope project to pin the install to one repository.

    shell
  4. 4

    Check it is installed

    Lists installed plugins. The entry should appear as enabled.

    shell

To remove it later, run claude plugin uninstall kagura-memory@kagura-memory-cloud.

Did setup work in Claude Code?

No setup reports for Claude Code yet. Self-reported by signed-in visitors. Copying a command is not counted, and most installs are never reported.

Attested installs: Fewer than 10 attested installs of this plugin in week 2026-W46, so no number is shown. Counted from opt-in HOL Guard install reports, once per machine and runtime, and never mixed with the answers above. How install reports work.

Loading reviews

Plugin manifest

{
  "name": "kagura-memory",
  "version": "0.173.2",
  "description": "Kagura Memory Cloud plugin — restore session context, recall & save durable project knowledge, setup help, an MCP smoke test, and tool-guardrail hooks",
  "author": {
    "name": "Kagura AI, Inc.",
    "url": "https://github.com/kagura-ai"
  },
  "homepage": "https://github.com/kagura-ai/memory-cloud",
  "repository": "https://github.com/kagura-ai/memory-cloud",
  "license": "MIT",
  "keywords": [
    "memory",
    "mcp",
    "ai-memory",
    "session-management",
    "knowledge-base"
  ],
  "commands": [
    "./claude-skills/"
  ],
  "hooks": "./claude-hooks/hooks.json",
  "userConfig": {
    "server_url": {
      "type": "string",
      "title": "Kagura Memory MCP URL",
      "description": "The MCP endpoint from .mcp.json — the whole URL, ending in /mcp or /mcp/w/<workspace-id>, never the site root. Example: https://<host>/mcp (or https://<host>/mcp/w/<workspace-id>). The hooks POST load_guardrails to exactly this URL, so a site root answers http 404/405 and no guardrail is ever delivered. Add ?guardrails=off to the .mcp.json URL itself (&guardrails=off when the URL already has a query, such as ?profile=core) so the server does not also send a guardrail digest. A kagura-mcp entry (kagura setup claude --profile) has no URL there: add --guardrails off to its args instead (kagura-mcp 0.39.0+). Adding context=<context-id> to the .mcp.json URL query (?context= or &context=) makes that context the default, so tools need no context_id and the session skips list_contexts. Keep that context= in server_url as well: the SessionStart hook then tells the model the default context, so no list_contexts call is made."
    },
    "api_key": {
      "type": "string",
      "title": "Kagura API key",
      "sensitive": true,
      "description": "A user API key (kagura_...). Agent-bound keys can read guardrails but not author them."
    },
    "context_id": {
      "type": "string",
      "title": "Guardrail context id",
      "description": "A UUID from list_contexts (the context's id, not its name), naming the context whose guardrails apply. One context for all projects in v1. /kagura-memory:setup picks it for you."
    },
    "max_action": {
      "type": "string",
      "title": "Strongest hook action",
      "default": "block",
      "options": [
        "block",
        "inform"
      ],
      "description": "block: a block guardrail denies the matching call once, with the memory as the reason. inform: never deny, only add context."
    }
  },
  "interface": {
    "displayName": "kagura-memory",
    "developerName": "kagura-ai"
  },
  "skills": "./",
  "holManifestOrigin": "repository",
  "registryIndexVersion": 9
}

Marketplace source

Repo URL
https://github.com/kagura-ai/memory-cloud
Marketplace path
.claude-plugin/marketplace.json
Source path
.
Install policy
Unspecified

Skills

1 SKILL.md file ship with this plugin. Preview, copy, or download each one, then install them with the Skills CLI.

Share
skills-cli
  • kagura-memory

    plugins/kagura-memory/skills/kagura-memory/SKILL.md

    Use Kagura Memory Cloud from Codex for long-term project memory. Trigger when the user asks to start or restore a session, recall past decisions or bug fixes, save session knowledge, remember a decision or troubleshooting note, review a context for outdated memories (maintain), inspect memory contexts, verify the kagura-memory MCP connection, or log in to it again (a Kagura tool failing with invalid_token or insufficient_scope, a new machine or workspace).

    Raw SKILL.md

File inventory

.agents/plugins/marketplace.json

file

394 bytes

d863bae267fcd559…

.claude-plugin/marketplace.json

file

530 bytes

882bbde5ea9965a6…

.claude-plugin/plugin.json

file

2,522 bytes

2344e76e76c3ead5…

.codex-plugin/plugin.json

plugin-manifest

2,409 bytes

eb2fbb9b4848ad67…

.github/dependabot.yml

file

2,264 bytes

63d6ad27a9b5bcde…

claude-hooks/hooks.json

file

2,373 bytes

13bc98c3fa97337b…

claude-skills/digest.md

file

4,575 bytes

df58ca3d0ffab5aa…

claude-skills/guide.md

file

17,415 bytes

7bdd2fc41cbf0f6b…

claude-skills/login.md

file

12,825 bytes

b2a94c20f4e21390…

claude-skills/maintain.md

file

8,181 bytes

c30e50336af558de…

claude-skills/recall.md

file

13,818 bytes

6c4173e62040c014…

claude-skills/remember.md

file

16,381 bytes

0e9b1ab36f84ae63…

claude-skills/session-start.md

file

12,705 bytes

d14d060b4826993e…

claude-skills/session-summary.md

file

19,215 bytes

3bc074c7e6ee1cc8…

claude-skills/setup.md

file

41,293 bytes

bab2a620750b6272…

claude-skills/smoke-test.md

file

61,010 bytes

84362f849140b551…

CLAUDE.md

file

4,214 bytes

44f95da1ed49f39b…

LICENSE

file

14,364 bytes

ba9e27599788ef87…

plugins/kagura-memory/skills/kagura-memory/SKILL.md

skill

48,675 bytes

519b17a756bcc8d7…

README.md

file

13,319 bytes

dc51f6ec02b885bd…

SECURITY.md

file

13,765 bytes

b3403f43366b1d73…

memory-cloud — Frequently asked questions

What is memory-cloud?
memory-cloud is an AI plugin in the HOL registry. MCP server: self-hosted persistent memory for AI agents, with remember and recall tools that keep project facts searchable across sessions.
How do I install memory-cloud?
Install memory-cloud in your harness: Claude Code — /plugin marketplace add kagura-ai/memory-cloud; MCP — git clone https://github.com/kagura-ai/memory-cloud; Any agent — npx [email protected] add kagura-ai/memory-cloud. Full step-by-step guidance is on the HOL plugin page.
How do I install memory-cloud in HOL Guard?
To install memory-cloud in HOL Guard, start with hol-guard plugin install kagura-ai/kagura-memory --apply. The complete step-by-step install guide for HOL Guard is on the HOL plugin page.
How do I install memory-cloud in Claude Code?
To install memory-cloud in Claude Code, start with /plugin marketplace add kagura-ai/memory-cloud. The complete step-by-step install guide for Claude Code is on the HOL plugin page.
How do I install memory-cloud in MCP?
To install memory-cloud in MCP, start with git clone https://github.com/kagura-ai/memory-cloud. The complete step-by-step install guide for MCP is on the HOL plugin page.
How is the Trust Score for memory-cloud computed?
memory-cloud has a HOL Trust Score of 35 out of 100. The score is a weighted sum of six components from a static scan of the source, capped for high-severity findings and adjusted by two disclosed policies. Memberships and payments are not inputs, and a scan is not a safety guarantee. Full breakdown: https://hol.org/registry/plugins/kagura-ai%2Fkagura-memory/trust
Who publishes memory-cloud?
memory-cloud is published by kagura-ai and listed on HOL.

Directory acceptance

Listed in 0 of 5 directory lists

Read from the list each directory publishes on GitHub. Being listed means the directory accepted the plugin; it is not a HOL security rating.

Lists last read . HOL re-reads them every day.

Submitting it? Check it against each directory's rules

Install signal

HOL cannot see every install. These are separate signals from different sources, so they are not added together. Counts under each floor show as <N.

Install commands issued here
<5
HOL Guard install commands issued from this listing, last 28 days.
Opt-in install reports
<10
Machines that reported installing it through HOL Guard in week 2026-W46. Reporting is opt-in, so most installs are not counted.
HOL Guard workspaces
<5
Workspaces that approved it or include it in an approved-plugin feed, as of today.

Reviews and similar plugins