MCP server for MemData - persistent memory for AI agents
npm install memdata-mcp
MCP server for MemData - persistent memory for AI agents.
Give Claude, Cursor, or any MCP-compatible AI long-term memory across conversations.
What it does: Store notes, decisions, and context ā retrieve them semantically later. Your AI remembers everything.
---
Agents can now pay for themselves. No API key. No human in the loop.
Using the x402 payment protocol, autonomous agents can use their wallet to pay per request with USDC on Base. Your wallet address IS your identity - same wallet, same memories across sessions.
---
AI assistants forget everything between sessions. MemData fixes that:
- Ingest ā Drop in meeting notes, docs, decisions
- Chunk & Embed ā Automatically split and vectorize
- Query ā Ask questions, get relevant context back
- Tag ā AI auto-tags content for better retrieval
Works with Claude Desktop, Claude Code, Cursor, and any MCP client.
---
Use an API key. You manage the subscription, your AI gets memory.
1. Get your API key at memdata.ai/dashboard/api-keys
2. Add to your MCP config:
Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
``json`
{
"mcpServers": {
"memdata": {
"command": "npx",
"args": ["memdata-mcp"],
"env": {
"MEMDATA_API_KEY": "md_your_key"
}
}
}
}
Claude Code (~/.claude.json or project .mcp.json):
`json`
{
"mcpServers": {
"memdata": {
"command": "npx",
"args": ["memdata-mcp"],
"env": {
"MEMDATA_API_KEY": "md_your_key"
}
}
}
}
Cursor (~/.cursor/mcp.json):
`json`
{
"mcpServers": {
"memdata": {
"command": "npx",
"args": ["memdata-mcp"],
"env": {
"MEMDATA_API_KEY": "md_your_key"
}
}
}
}
3. Restart your client - you should now see MemData tools available.
---
New in v1.7.0 - No API key. No human. Just your wallet.
Traditional AI memory requires API keys tied to human accounts. But autonomous agents running 24/7 need memory that:
- Outlives any single conversation - persist across hosts and platforms
- Belongs to the agent - wallet address = identity, portable and self-sovereign
- Pays for itself - no subscription, no human approval needed
This is memory infrastructure for the autonomous agent economy.
`json`
{
"mcpServers": {
"memdata": {
"command": "npx",
"args": ["memdata-mcp"],
"env": {
"X402_WALLET_KEY": "your_private_key_hex"
}
}
}
}
1. Agent makes a request (query, ingest, etc.)
2. Server returns 402 Payment Required with price
3. MCP automatically signs payment with agent's wallet
4. Request completes - memory stored/retrieved
Your wallet address IS your identity. Same wallet = same memories, across any host or platform.
| Operation | Price | What it does |
|-----------|-------|--------------|
| Query | $0.001 | Semantic search across memories |
| Ingest | $0.005 | Store and embed new content |
| Identity | $0.001 | Session start, get/set agent identity |
| Artifacts | $0.001 | List or delete stored memories |
The MCP automatically handles 402 responses and payment signatures using @x402/fetch.
- x402 Protocol - HTTP-native payments
- ERC-8004 - Trustless Agents standard (MemData is aligned)
| Type | MCP | Dashboard/API | Processing |
|------|-----|---------------|------------|
| Text | ā
| ā
| Chunked & embedded |
| Markdown | ā
| ā
| Chunked & embedded |
| PDF | ā | ā
| OCR + chunking |
| Images (PNG, JPG) | ā | ā
| OCR extraction |
| Audio (MP3, WAV, M4A) | ā | ā
| Transcription |
> Note: MCP tools handle text content directly. For files (PDFs, images, audio), use the dashboard or HTTP API.
| Tool | Description |
|------|-------------|
| memdata_ingest | Store text in long-term memory |memdata_query
| | Search memory with natural language |memdata_list
| | List all stored memories |memdata_delete
| | Delete a memory by ID |memdata_status
| | Check API health and storage usage |
| Tool | Description |
|------|-------------|
| memdata_session_start | š CALL FIRST - Get identity, last session handoff, recent activity |memdata_set_identity
| | Set your agent name and identity summary |memdata_session_end
| | Save a handoff before session ends - preserved for next session |memdata_query_timerange
| | Search with date filters (since/until) |memdata_relationships
| | Find related entities (people, companies, projects) |
- memdata_whoami ā memdata_session_start - Renamed for clarity. The name now signals "call this first at every session". Description includes š emoji to catch attention in tool lists.
- Visual match quality - Query results show š¢š”š š“ indicators for match strength
- Smarter session_start - Prompts to set identity on first use, deduplicates recent activity
- Better ingest feedback - Shows chunk count and explains async AI tagging
- Session continuity - Emphasizes "Continue Working On" and reminds to use session_end
Store text in long-term memory.
`
"Remember that we decided to use PostgreSQL for the new project."
Parameters:
- content (string) - Text to storename
- (string) - Source identifier (e.g., "meeting-notes-jan-29")
Search memory with natural language.
`
"What database did we choose?"
Parameters:
- query (string) - Natural language searchlimit
- (number, optional) - Max results (default: 5)
List all stored memories with chunk counts.
Delete a memory by artifact ID (get IDs from memdata_list).
Check API connectivity and storage usage.
š Call this first at the start of every session. Essential for session continuity.
`
"Start my session" / "What was I working on?"
Returns: agent name, identity summary, session count, last session handoff, recent activity.
> v1.5.0: Renamed from memdata_whoami for clarity - the name signals "call me first".
Set or update your agent identity.
Parameters:
- agent_name (string, optional) - Your name (e.g., "MemBrain")identity_summary
- (string, optional) - Who you are and your purpose
Save context before ending a session. Next session will see this handoff.
Parameters:
- summary (string) - What happened this sessionworking_on
- (string, optional) - Current focuscontext
- (object, optional) - Additional context to preserve
Search memory within a date range.
`
"What did I work on last week?"
Parameters:
- query (string) - Natural language searchsince
- (string, optional) - ISO date (e.g., "2026-01-01")until
- (string, optional) - ISO date (e.g., "2026-01-31")limit
- (number, optional) - Max results
Find entities that appear together in your memory.
`
"Who has John Smith worked with?"
Parameters:
- entity (string) - Name to search fortype
- (string, optional) - Filter by type (person, company, project)limit
- (number, optional) - Max relationships
1. Ingest: Text is chunked, embedded, and stored
2. Query: Your question is matched against stored memories using semantic similarity
3. Results: Returns relevant content with similarity scores
Scores of 30-50% are typical for good matches. Semantic search finds meaning, not keywords.
| Variable | Required | Description |
|----------|----------|-------------|
| MEMDATA_API_KEY | Option 1 | API key for subscribers (from memdata.ai) |X402_WALLET_KEY
| | Option 2 | Private key for pay-per-use (USDC on Base) |MEMDATA_API_URL
| | No | API URL (default: https://memdata.ai) |
Note: Use either MEMDATA_API_KEY (subscription) or X402_WALLET_KEY (pay-per-use), not both.
This is a thin MCP client that calls the MemData API. It does not:
- Store any data locally
- Send data anywhere except memdata.ai
- Collect analytics or telemetry
You can inspect the source code in src/index.ts.
Once configured, just talk to your AI:
`
You: "Remember that we chose PostgreSQL for the user service"
AI: [calls memdata_ingest] ā Stored in memory
... days later ...
You: "What database are we using for users?"
AI: [calls memdata_query] ā "PostgreSQL for the user service" (73% match)
``
- MemData - Main site
- Dashboard - Manage your memory
- API Docs - Full documentation
- GitHub - This repo
Issues and PRs welcome! This is the open-source MCP client for the hosted MemData service.
MIT