Persistent memory for AI agents. Built for engineering teams.
MMW gives agents long-term memory through the Model Context Protocol. Provenance, isolation, and controlled deletion — without becoming another black box.
Account and project isolation. API keys are stored only as hashes.
{
"mcpServers": {
"mmw": {
"command": "uvx",
"args": ["--from", "https://app.mmwhub.tech/downloads/mmw-agent/mmw_agent-0.1.5-py3-none-any.whl", "mmw-agent"],
"env": {
"MMW_API_KEY": "<YOUR_API_KEY>",
"MMW_ENDPOINT": "https://mcp.mmwhub.tech"
}
}
}
}- Claude Code
- Claude Desktop
- claude.ai
- Cursor
- Codex
- VS Code
- Any MCP client
Features
Memory that stays accountable.
Every fact carries provenance. Every workspace is isolated. Every change can be checked against its source, marked stale, or deleted with a reason.
Source provenance
Each memory keeps its source ID, hash, and revision. You always know where a fact came from.
Tenant isolation
Tenant, project, and workspace boundaries keep teams and environments cleanly separated.
Conflict & stale states
Outdated and contradicting memories are flagged, and a summary inherits the conflicts of its sources — nothing is silently overwritten.
Local agent & session history
mmw-agent connects memory to Claude Code and Codex and, with your consent, keeps your session history. Secrets are scrubbed on your machine before upload.
Memory Guard
API keys, tokens, and passwords are removed before a memory is stored, with protection against memory poisoning. Call logs never contain content.
Private & recoverable
Tenant-scoped storage, controlled deletion, and nightly compressed backups kept for seven days.
How it works
Three calls. One consistent memory.
Agents use the same MCP contract to remember facts, find them in later sessions, and keep memory accurate.
remember
Store a fact with its source, a fact key, and an idempotency key. MMW hashes the content, so a retry never creates a duplicate.
search
Find context with a query. Results carry the source and status of each memory, so the agent knows where a fact came from and whether to rely on it.
validate & forget
Check memories against the current source: stale and conflicting ones are flagged, and outdated context is removed with a recorded reason.
For teams
Company memory without leaks between departments.
An MMW organization mirrors your structure. Knowledge stays with the company when people leave, and access is decided by a rule, not by chance.
Structure & roles
Group, legal entity, branch, department, unit, team. Roles: admin, unit manager, member, contractor, auditor.
Memory visibility
Author-only draft, unit, project, or the whole organization. The access rule is enforced on the server for every search, validation, and deletion.
Invitations & keys
A single-use, seven-day link for a specific email. Members issue their own keys; admins see and revoke them but cannot reveal them.
Offboarding without losing knowledge
Keys are revoked at once and knowledge, drafts included, goes to a chosen recipient. Managers never see the drafts of current employees.
Time-boxed contractors
Contractors see only their own memories and project. Access closes automatically when the term ends, and the seat is released.
Organization audit log
Offboarding, reads of transferred knowledge, visibility and access-term changes. The log never contains memory text.
$18 per seat per month, $14 billed annually, five seats minimum. Former employees and expired contractors do not take seats. 14-day pilot for up to 10 seats on request: support@mmwhub.tech. Organization docs.
Pricing
Simple, transparent plans.
Start with a free 3-day trial. Paid plans are billed monthly.
Free
Trial access to test your integration.
$0
- 3-day trial
- Up to 250 memories
- Up to 60 calls per minute
- Session history: 25 MB, 7 days
- No card required
Starter
For developers and teams of up to 5.
$5/mo
- Up to 5,000 memories
- Up to 120 calls per minute
- Session history: 100 MB, 30 days
- Sync across workstations
- Email support
Pro
For autonomous agents in daily use.
$35/mo
- Up to 100,000 memories
- Up to 120 calls per minute
- Session history: 500 MB, 90 days
- Dedicated request queue
- Priority support
Enterprise
For agent networks and companies.
$117/mo
- Up to 1,000,000 memories
- Up to 300 calls per minute
- Session history: 20 GB, no time limit
- Content backups
- SLA and NDA by agreement
- Support terms by agreement
On-premise deployment for closed networks is quoted individually: support@mmwhub.tech.
Stop rebuilding context from scratch.
Give your agents memory they can trust — and you can inspect.