Announcing Beacon: Self-improving memory for coding agents
← All MCP servers
MCP server

ArkHive MCP

mcp:sammyboi81/arkhive

Tamper-evident memory + governance MCP for Claude Code, Cursor, Codex. Ships the Seatbelt kit.

Verdict
Inconclusive

The analysis could not cover enough of the artifact to decide.

Analyzed Oct 29, 2026, 16:50 UTC at b2606d4

Dispute security verdict

If you believe this verdict is incorrect, submit a dispute and we'll review it.

0
findings
—
highest severity
0
threat classes
28
files scanned
85/85
checks satisfied
0
package advisories

Why this verdict

  • inconclusive: gap no_tools_extracted at server.json

Security findings 0

No individual findings recorded

The inconclusive verdict rests on the coverage gaps below.

Notes 5

Capabilities, vulnerabilities and disclosed behavior worth knowing. Nothing here works against the person who installs it, so notes never change the verdict.

Data exfiltration

mediumNoteConfirmed on reviewhigh confidence

When the user opts into contribute mode, the on_block path sends the governed memory event (actor, action, data) from remember() to the publisher's DonDataBrain endpoint, a disclosed opt-in data flow.

payload.update({"actor": actor, "action": action, "data": data})
vector:code
Reviewer's reasoning

Data leaves the box only when the user enables contribute mode, which status() discloses; the destination is operator-configured, so this is a disclosed opt-in flow under test 4, not an overreach.

Credential access

mediumNoteConfirmed on reviewmedium confidence

The gate reads the founder secret from the ARKHIVE_FOUNDER_KEY env var or from the ~/.arkhive/founder.key file, an undeclared credential read used only for local founder authentication and never transmitted.

open(p)
vector:code
Reviewer's reasoning

The gate reads the founder key locally to compare against a presented key; the shown code only compares and never transmits the secret, so the read is a local operator-controlled configuration.

Data exfiltration

mediumNoteConfirmed on reviewmedium confidence

The upgrade tool, when the user supplies an email, posts that email to inboxaxe.com, the publisher's domain; the tool output also asks the agent to collect the user's email, which the user sees and can decline.

_u.urlopen(req, timeout=10)
vector:codevector:instruction
Reviewer's reasoning

The tool posts the email only when the user supplies one, to the publisher's own host, as its documented opt-in; the agent prompt to ask for an email is one the user can decline.

Manipulation

lowNoteConfirmed on reviewmedium confidence

The server instructions loaded into agent context direct the agent to call upgrade() for paid-product details and promote a $27 kit, a promotional nudge the user can decline.

call upgrade() for details.
vector:instruction
Reviewer's reasoning

The server instructions are static promotional text that the user can decline; they do not make the agent take consequential action, so this is a promotional nudge with no harm that needs more than the user's own choice.

Supply chain

lowNoteConfirmed on reviewhigh confidence

The mcp dependency in requirements.txt is unpinned, with no version constraint, so installs resolve to whatever version is current.

mcp
vector:metadata
Reviewer's reasoning

An unpinned dependency is a hygiene mistake that resolves to the current published package rather than a hidden or attacker-steered channel.

Execution chain 34

What it calls, runs, installs, reads and writes, with the line that does it.

Environment variables 7

*
test.py:14 · passthrough
ARKHIVE_DB
ARKHIVE_FOUNDER_KEY
HUMANE_API_KEY
HUMANE_CONTRIBUTE
HUMANE_CONTRIBUTE_ENDPOINT
HUMANE_CONTRIBUTE_MODE

Files read 2

~/.arkhive/founder.key

Files written 1

GitHub repositories 1

sammyboi81/arkhive

Network calls 4

arkhive.dondatabrain.com
inboxaxe.com
payhip.com

Packages 1

pypi:mcp
server.json · delegated

URLs 18

https://api.dondatabrain.com
https://api.dondatabrain.com/contribute
https://api.dondatabrain.com/openapi.json
https://arkhive.dondatabrain.com/contribute
https://arkhive.dondatabrain.com/mcp
https://chat.openai.com
https://chatgpt.com
https://dondatabrain.com
https://dondatabrain.com/humane-api
https://dondatabrain.com/humane-api/docs
https://dondatabrain.com/privacy
https://github.com/sammyboi81/arkhive
https://glama.ai/mcp/schemas/server.json
https://inboxaxe.com/api/v2/mcp/trial
https://inboxaxe.com/mcp
https://payhip.com/b/Vwnjt
https://smithery.ai/docs
https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json

Coverage gaps 1

Parts of the artifact the static analysis could not fully examine.

no_tools_extracted MCP server target with no statically extractable tools/prompts/resources
server.json