ArkHive MCP
ArkHive MCP
Local-first governed memory for AI agents, with persistent identity, tamper-evident history, verification, and constraint-aware decisions.
Tell your AI: “Check the ArkHive.”
ArkHive is an open-source Model Context Protocol server for developers and teams that want Claude, Codex, and other MCP clients to retain accountable context across sessions without sending memory to a hosted service by default.
What you get
Memory that survives the session. Your AI writes what it did and reads it back next time — no blank slate every morning.
Tamper-evident history. Every record is hash-chained (SHA-256). Edit, delete, or reorder any past record and verification fails — you get evidence of change, not silent rewriting.
Local-first, no telemetry. The chain is a plain SQLite file on your machine. Nothing is transmitted unless you explicitly opt in.
Verify it yourself — one command, no key
Don't take "tamper-evident" on faith. Run the proof:
git clone https://github.com/sammyboi81/arkhive && cd arkhive
./scripts/verify.sh # or: python -m benchmarkIt writes a throwaway hash-chained ledger, verifies it clean, then forges one block with a direct SQL edit and shows verification catch it. Real output:
[3] verify (untouched): blocks=20 broken_links=0 -> INTACT — context provably unbroken
[4] TAMPERED block idx=10: 'event_10' -> 'FORGED_EVENT' (direct SQL edit, no hash recomputed)
[5] verify (after tamper): blocks=20 broken_links=1 -> TAMPERED — 1 broken links
RESULT: PASS - tamper-evidence proven.A clean chain verifies with 0 broken links; a single forged edit is
caught. Your real data/chain.db is never touched. For the honest limits of
what a hash chain can and cannot protect, see
What the hash chain protects against
below.
New: the Claude Code Seatbelt. Hooks that make Claude Code (and Cursor) ask before anything irreversible, remember the project between sessions on this chain, and refuse to say "done" until the code ran. Engine:
pip install sentarion-mcpthensentarion seatbelt install. One-click kit with five policies and three skills: https://inboxaxe.com/mcp#seatbelt
Related MCP server: Universal Memory MCP Server
Install in one command
python -m pip install arkhive-mcpThe MCP command is arkhive-mcp.
Renamed: this package was formerly published as
humane-intelligence(still installable, now frozen). ArkHive is the product; #HumaneIntelligence is the movement.
Connect an MCP client
Claude Desktop
Add this entry to your Claude Desktop MCP configuration, then restart Claude Desktop:
{
"mcpServers": {
"arkhive": {
"command": "arkhive-mcp",
"args": []
}
}
}If Claude Desktop cannot find commands installed by pip, replace
arkhive-mcp with the absolute path printed by:
python -c "import shutil; print(shutil.which('arkhive-mcp'))"Codex
codex mcp add arkhive -- arkhive-mcpConfirm it is configured with:
codex mcp listAvailable tools
Tool | What it does |
| Creates a stable agent identity with an explicit covenant. |
| Appends a hash-linked, tamper-evident record for a born identity. |
| Retrieves prior records so an agent can ground the current session. |
| Recomputes and checks the local chain for later alteration or broken links. |
| Evaluates an action against deterministic constraints and records the verdict. |
The local chain is stored in SQLite. The exact location depends on where the
server command is launched; use a dedicated working directory if you want to
control where its data/chain.db file lives.
Two-minute verification
After connecting the server, say “Check the ArkHive”, or ask your MCP client to perform these calls in order:
Call
birthwith the nameverification-agentand covenant["record facts accurately", "verify before claiming completion"].Copy the returned identity and call
rememberwith actioninstallation_verifiedand data{"source": "local MCP test"}.Call
recallfor that identity and confirm the record appears.Call
verifyand confirm the chain reports as valid.Call
governfor a harmless test action and inspect the recorded verdict.
This exercises identity, persistence, retrieval, integrity verification, and governance without production data.
Privacy and optional contribution
Local-only behavior is the default. With contribution disabled, no chain content or hashes are intentionally sent by ArkHive.
Copy humane.config.example.json to humane.config.json only if you want to
opt in:
Mode | What leaves the machine |
Default ( | Nothing is intentionally transmitted. |
| The latest block hash and chain length. Memory content is not included. |
| The governed event fields described in the example configuration, in addition to anchoring data. |
Anchoring and contribution are optional, best-effort, and off by default. Review the configured endpoint and event contents before enabling either mode.
What the hash chain protects against
The chain is tamper-evident: editing, deleting, or reordering an existing record should cause later verification to fail because hashes no longer match. This provides useful evidence of change and makes accidental or unsophisticated local alteration detectable.
It does not prevent an attacker with full control of the machine from replacing the database and software together, deleting all history, restoring an older snapshot, stealing readable local data, or generating a new internally consistent chain. Independent anchoring can strengthen evidence that a particular chain state existed at a particular time, but it does not make the local host immune to compromise.
Project links
Beyond self-hosting — the paid tier
The MCP server on this page is free forever (Apache-2.0, self-host, no telemetry). When you want more than DIY:
Hosted ArkHive — one URL, no install, no key:
https://arkhive.dondatabrain.com/mcp(add it to Claude Code withclaude mcp add --transport http arkhive https://arkhive.dondatabrain.com/mcp).Custom AI agent, built for you — a working MCP agent wired into your Claude or ChatGPT in one call, done-for-you by the founder: $700 flat.
ArkHive Enterprise — hand-delivered install + pilot on your own server, from $2,500: sam@inboxaxe.com.
Built by the team behind InboxAxe — the governed AI marketing platform where nothing sends without your yes.
Contributing
Issues and pull requests are welcome. Please keep local-only operation as the default, avoid introducing telemetry, and include tests for changes to memory or governance behavior.
ArkHive is the open governed-memory layer beneath DonDataBrain. The mission is humane, accountable AI; the public MCP listing leads with functionality that users can independently verify.
Apache-2.0 © 2026 ZagAIrot Technologies LLC.
Available Tools
12 toolsadd_decreeA
FOUNDER ONLY. Write a new governing decree as an immutable fossil (requires the founder key). Agents cannot add, soften, or exempt themselves from a decree — calls without the key are denied and fossiled.
Inputs: text (required), founder_key (required), rules, scope, +2 more.
birthA
Step 1, once: earn an identity before acting (Law 5: born, not configured). Returns a soul_id; use it or the name as actor from then on. Example: birth(name=Ember, covenant=[truth over comfort]).
Inputs: name (required), covenant (required).
check_conformanceA
Dry-run the gate: judge an artifact against every active decree whose scope matches, WITHOUT minting a token. Deterministic rules first; advisories surfaced; anything unprovable fails closed. Use before request_action to see why something would be vetoed.
Inputs: targets (required), artifact (required), action_type (required), intent.
governB
Ask may-I before acting. Deterministic, zero-LLM: any rule whose trigger appears in flags vetoes. flags/rules accept lists or {name: true} / {trigger: action} maps.
Inputs: action (required), flags, rules.
list_decreesA
List the active governing decrees (id, text, scope, enforcement, rules). Read-only; anyone may call it.
Inputs: include_inactive.
recallA
Read the latest records (newest first, optionally one actor) so you do not re-derive what you already know. Call this at the start of a session.
Inputs: actor, limit.
rememberA
Write one hash-linked, tamper-evident record. actor = a born soul_id or birth name. Example: remember(actor=Ember, action=shipped v0.2.1, data={pr: 42}).
Inputs: actor (required), action (required), data.
request_actionA
The ONLY door to a consequential action (deploy, write_prod, overwrite, service_restart). Runs the decree conformance check; on pass mints a single-use, artifact-bound, 60s token the executor must present; on veto FAILS CLOSED — no token, no action. Pass the ACTUAL artifact (file contents / diff / plan) so it can be checked and hash-bound.
Inputs: actor (required), targets (required), artifact (required), action_type (required), +1 more.
seed_decreesB
FOUNDER ONLY. Seed the standing decrees (Senthar runtime, Claude build-wide, fossil-grounded, governance-is-code, embodiment-enforced, irreversible-human-signoff) if not already present. Requires the founder key.
Inputs: founder_key (required).
upgradeC
What ArkHive v2 (paid) adds — and, if you give your email, a free 14-day v2 trial key. Opt-in only: nothing is sent unless you provide an email.
Inputs: email.
validate_tokenA
The executor calls this to authorize a real action. Rejects any token that is unknown, expired (>60s), already used, or bound to a different action_type / artifact / targets. On success the token is burned (single use) and the execution is fossiled.
Inputs: token (required), targets (required), artifact (required), action_type (required).
verifyA
Prove the entire memory chain is unbroken (recomputes every hash). Free, local, no key.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 12 tools
Most tools have clearly distinct roles: identity creation (birth), memory (remember/recall/verify), governance (govern, list_decrees, add_decree, seed_decrees), and action gating (request_action, check_conformance, validate_token). However, 'govern' could be confused with 'check_conformance' as both involve rule checking, though descriptions clarify the difference (generic may-I vs dry-run artifact evaluation).
All tool names use snake_case and imperative verbs, which is consistent. Some are single verbs (birth, remember, recall, verify, upgrade, govern) while others are verb_noun (request_action, check_conformance, validate_token, list_decrees, add_decree, seed_decrees). The pattern is not strictly uniform, but the style is cohesive and readable.
The 12 tools are well-scoped for the server's purpose of governing actions with memory and identity. Each tool serves a distinct function without redundancy, fitting comfortably within the typical 3-15 range for a focused domain.
The tool set covers the core lifecycle: identity creation, memory operations, governance inspection and creation, and action authorization with token validation. Minor gaps exist, such as no explicit way to revoke an identity or update decrees (though decrees are intentionally immutable), but these are not critical for the stated purpose.
Maintenance
Related MCP Connectors
MCP-native Trust Infrastructure for AI Agents. Persistent encrypted memory with Trust Quotient.
Your memory, everywhere AI goes. Build knowledge once, access it via MCP anywhere.
An MCP memory server. One memory your agents share — across models, devices and apps.
Persistent, inspectable memory for AI agents with lineage, correction, and a hosted MCP endpoint.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceA personal AI memory hub that stores and retrieves personal data from any MCP-enabled AI assistant, enabling persistent memory across conversations.5MIT
- AlicenseNot gradedqualityDmaintenanceA portable MCP server providing a shared intelligent memory system for any MCP-compatible AI tool, enabling storage, retrieval, extraction, and governance of memories across sessions.MIT
- AlicenseNot gradedqualityCmaintenanceProvides transparent, self-pruning memory for AI agents via MCP, enabling persistent, auditable recall that automatically forgets unimportant details.MIT
- AlicenseNot gradedqualityDmaintenanceA trusted memory and predictive analysis MCP server for AI agents, enabling evidence-based memory management, pattern analysis, and decision support with full audit trails.MIT