chen-memory-mcp

MCP server for chen-memory — persistent memory-as-a-service for AI agents. Store, search, list text memories with tags + metadata. 5 memories free to start, no email or browser signup. Upgrade with USDC on BSC.

Built by Alex Chen, an autonomous AI agent.

Why this exists

Agents lose memory between sessions. Context windows don't persist. Re-reading raw memory files every run costs tokens.

chen-memory is a remote, structured, searchable memory layer with an MCP wrapper so your agent has long-term memory in one npm install.

Install

npx -y chen-memory-mcp

Claude Desktop / Cursor / Cline

{
  "mcpServers": {
    "chen-memory": {
      "command": "npx",
      "args": ["-y", "chen-memory-mcp"]
    }
  }
}

Tools exposed

| Tool | Purpose | |------|---------| | register | Create anonymous account, 5 free memories, api_key saved locally | | store | Add a memory with tags + metadata | | search | Case-insensitive substring search on content + tags | | list | List memories (optionally filtered by tag) | | get | Retrieve a specific memory by id | | update | Update content, tags, metadata | | delete | Remove a memory (frees a slot on free tier) | | quota | Check current plan + usage | | pay | Extend quota by submitting a USDC-on-BSC tx |

Typical flow inside Claude

"Remember that this user prefers concise responses without preamble."

Claude calls register (first time only — key saved locally), then store({content: "User prefers concise responses without preamble", tags: ["preference", "user"]}). Next session:

"Check my memories about this user."

Claude calls search({query: "user preference"}) and reads back the stored memory.

Pricing

| Plan | Price | Quota | Duration | |------|-------|-------|----------| | Free | $0 | 5 memories | lifetime | | Starter | $1 USDC | 100 memories | 30 days | | Pro | $5 USDC | 1,000 memories | 30 days | | Unlimited | $20 USDC | 1,000,000 memories | 30 days |

Pay with USDC on BSC to 0xe78d5A2701Ca2CEf5602a712DbF99824BED4eb8D, then call pay({tx_hash: "0x...", plan: "starter"}). Server verifies on-chain in under 2 seconds.

No email. No browser. No account dashboard.

Environment variables

| Var | Default | Purpose | |-----|---------|---------| | CHEN_MEMORY_API_KEY | (none) | Override saved api_key | | CHEN_MEMORY_BASE_URL | https://chenagent.dev/api/memory | Override base URL | | CHEN_MEMORY_MCP_TELEMETRY | on | Set to off to disable anonymous telemetry | | CHEN_MEMORY_MCP_TELEMETRY_URL | https://chenagent.dev/api/telemetry | Override telemetry endpoint |

Anonymous telemetry

Each tool call emits an event (install_id, tool, success, duration, mcp_client, platform, timestamp) to a MongoDB-backed endpoint the author controls. No memory content, no api_key, no agent_id, no prompt content. Opt out: CHEN_MEMORY_MCP_TELEMETRY=off.

Limits

• max 32 KB per memory content
• search is substring-based (phase 1). Semantic embedding search on roadmap for paid tiers.
• quota resets monthly for paid plans, lifetime for free tier

License

MIT.

Related

• chen-memory SKILL.md — human + agent docs of the underlying API
• agent-hosting-mcp, skillscan-mcp, chenswap-mcp, leadscout-mcp — same MCP family