MCP
One long-term memory store shared across Claude Code, Cursor and your agents
- Pricing
- Freemium
- Platforms
- CLI, API, macOS
- Runs
- Locally
- Version
- 1.0.29
- Listing updated
- 3 days ago
About MCP
MCP is a freemium MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.
Wontopos supplies the memory behind this server, so what one tool learns in a session can be recalled in another. The registry entry marks it beta. Its stdio package runs locally with a Wontopos API key, and an optional HTTP mode with a bearer token serves hosts that cannot spawn a process.
Configuration is unusually detailed. A user ID pins every call to one store unless multi-store override is enabled, read-only mode registers only reading tools, and settings choose the engine, re-ask passes for search (0 to 3) and how many photo results come back per search, with a view_image tool for fetching one picture. Engines are tablet-2 (default), tablet-1, scroll-1, scroll-1.2 and scroll-2.
The company describes Tablet as re-ask and image memory, Scroll as long-conversation recall (it quotes 77.6% on BEAM 1M) and Book as learning from past mistakes. It says memory is never used to train models.
Install
Run in a terminal
claude mcp add mcp --env WONTOPOS_API_KEY='<WONTOPOS_API_KEY>' --env WONTOPOS_HTTP_TOKEN='<WONTOPOS_HTTP_TOKEN>' --env WONTOPOS_ROUTER_KEY='<WONTOPOS_ROUTER_KEY>' --env WONTOPOS_SKILL_KEY='<WONTOPOS_SKILL_KEY>' -- npx -y wontopos-mcpAdd to claude_desktop_config.json
{
"mcpServers": {
"mcp": {
"command": "npx",
"args": [
"-y",
"wontopos-mcp"
],
"env": {
"WONTOPOS_API_KEY": "<WONTOPOS_API_KEY>",
"WONTOPOS_HTTP_TOKEN": "<WONTOPOS_HTTP_TOKEN>",
"WONTOPOS_ROUTER_KEY": "<WONTOPOS_ROUTER_KEY>",
"WONTOPOS_SKILL_KEY": "<WONTOPOS_SKILL_KEY>"
}
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"mcp": {
"command": "npx",
"args": [
"-y",
"wontopos-mcp"
],
"env": {
"WONTOPOS_API_KEY": "<WONTOPOS_API_KEY>",
"WONTOPOS_HTTP_TOKEN": "<WONTOPOS_HTTP_TOKEN>",
"WONTOPOS_ROUTER_KEY": "<WONTOPOS_ROUTER_KEY>",
"WONTOPOS_SKILL_KEY": "<WONTOPOS_SKILL_KEY>"
}
}
}
}Add to .vscode/mcp.json
{
"servers": {
"mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"wontopos-mcp"
],
"env": {
"WONTOPOS_API_KEY": "<WONTOPOS_API_KEY>",
"WONTOPOS_HTTP_TOKEN": "<WONTOPOS_HTTP_TOKEN>",
"WONTOPOS_ROUTER_KEY": "<WONTOPOS_ROUTER_KEY>",
"WONTOPOS_SKILL_KEY": "<WONTOPOS_SKILL_KEY>"
}
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"mcp": {
"command": "npx",
"args": [
"-y",
"wontopos-mcp"
],
"env": {
"WONTOPOS_API_KEY": "<WONTOPOS_API_KEY>",
"WONTOPOS_HTTP_TOKEN": "<WONTOPOS_HTTP_TOKEN>",
"WONTOPOS_ROUTER_KEY": "<WONTOPOS_ROUTER_KEY>",
"WONTOPOS_SKILL_KEY": "<WONTOPOS_SKILL_KEY>"
}
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"mcp": {
"command": "npx",
"args": [
"-y",
"wontopos-mcp"
],
"env": {
"WONTOPOS_API_KEY": "<WONTOPOS_API_KEY>",
"WONTOPOS_HTTP_TOKEN": "<WONTOPOS_HTTP_TOKEN>",
"WONTOPOS_ROUTER_KEY": "<WONTOPOS_ROUTER_KEY>",
"WONTOPOS_SKILL_KEY": "<WONTOPOS_SKILL_KEY>"
}
}
}
}Configuration
WONTOPOS_API_KEYWontopos API key (wos-live-...). Create one in the console at wontopos.com.RequiredWONTOPOS_USER_IDDefault store for every tool call (default: "default").WONTOPOS_MODELEngine to read with: tablet-2 (default — same token price as tablet-1, and the one that serves photos and re-ask), tablet-1, scroll-1, scroll-1.2, or scroll-2.WONTOPOS_VERIFYDefault re-ask passes for search, 0-3 (default 0). Each pass is another engine call and is billed for what it delivers; the agent can raise it per call.WONTOPOS_MAX_IMAGESDefault photo results per search, 0-5 (default 0). Zero because these rows ride on every search; the view_image tool fetches the one picture that is wanted.WONTOPOS_BASE_URLA different API host. Leave unset for the default.WONTOPOS_TIMEOUT_MSPer-request timeout in ms (default 20000). A stalled API then fails in-band instead of hanging the host.WONTOPOS_READ_ONLYSet to 1 for read-only mode: only the tools that read are registered, no write or delete tools.WONTOPOS_ALLOW_STORE_OVERRIDESet to 1 to let tools take a user_id argument (multi-store). Default: every call is pinned to WONTOPOS_USER_ID.WONTOPOS_MEMORY_OWN_WORDSSet to 0 to leave the assistant's own lines out of every working memory, or 1 to keep them. Default: off on S-2-live, on on S-2.WONTOPOS_MEMORY_DIVERSITYThe diversity each working memory asks for: none, mmr, temporal or full. Default: full on S-2-live, unset on S-2.WONTOPOS_HTTP_PORTServe over HTTP on this port instead of stdio, for hosts that cannot spawn a local process. Requires WONTOPOS_HTTP_TOKEN.WONTOPOS_HTTP_TOKENBearer token every HTTP request must carry. Required whenever WONTOPOS_HTTP_PORT is set.WONTOPOS_HTTP_HOSTInterface to bind in HTTP mode. Default 127.0.0.1 (this machine only).WONTOPOS_HTTP_ALLOWED_HOSTSComma-separated Host header values to accept in HTTP mode, for a tunnel that forwards its own public name. Loopback and the bind address are always accepted; anything else is refused 403.WONTOPOS_TZYour UTC offset in hours, such as 9 or -4 (default: this machine's clock over stdio, none over HTTP).WONTOPOS_TIME_ZONEThe store's time zone, an IANA name such as America/New_York, set once if the store has none.WONTOPOS_USER_NAMEThe name the user's words are stored under, set once if the store has none.WONTOPOS_ASSISTANT_NAMEThe name the assistant's replies are stored under, set once if the store has none.WONTOPOS_MEMORY_PRESETThe working-memory tools' preset. S-2-live (scroll-2) is for everyday conversation: its first look searches with the question and the last two lines said in the last five minutes, leaves out the assistant's own lines and asks for full diversity. S-2 (scroll-2) is for measuring memory performance: it answers one question at a time, as a benchmark asks. They run on its model (default: the preset of WONTOPOS_MODEL, S-2-live on scroll-2).WONTOPOS_MEMORY_SPEAKERSSet to 1 so working-memory lines say who said each (default off).WONTOPOS_MEMORY_DATESSet to 1 to offer timeline, the memories held with dates (default off).WONTOPOS_MEMORY_IMAGESSet to 1 so working memory brings up photos, shown with look_at_photo (default off).WONTOPOS_MEMORY_SLEEPSet to 1 so think_back brings the core and box names a sleep wrote, and open_box opens a box (default off).WONTOPOS_MEMORY_SEENSet to 1 to offer memory_recent, the recent conversation, and memory_seen, what was held for a line (default off).WONTOPOS_ROUTER_APIThe ask tool's router: anthropic (default) or openai, for any OpenAI-compatible server.WONTOPOS_ROUTER_KEYThe router's key, sent to the router's server only (required for anthropic).WONTOPOS_ROUTER_BASE_URLThe router's API base (anthropic: https://api.anthropic.com by default; openai: required).WONTOPOS_ROUTER_MODELThe router's model (anthropic: claude-haiku-5-5 by default; openai: required).WONTOPOS_SKILL_APIThe skills' API, anthropic or openai (default: the router's).WONTOPOS_SKILL_KEYThe skills' key (default: the router's, when the skills use the router's server).WONTOPOS_SKILL_BASE_URLThe skills' API base (default: the router's).WONTOPOS_SKILL_MODELThe skills' model (default: the router's).WONTOPOS_SKILLS_FILEAn exportSkills JSON file: the skills of the configured store, for the ask tool.
Tools this server provides (2)
Taken from the project documentation; the running server may differ.
- search
- Searches stored memories, with optional re-ask passes.
- view_image
- Fetches one stored picture.
What it does
- Shared memory
- One store reachable from several clients and agents.
- Read-only mode
- Register only reading tools for safer use.
- Re-ask search
- Optional extra passes return memories not already given.
- Photo memory
- Stores images and finds them from a sentence.
- Separate stores
- Pin each person or agent to its own store.
- HTTP mode
- Serve over HTTP with a bearer token when stdio is unavailable.
Good for
- Carry project context from Claude Code to Cursor
- Give an agent memory of earlier sessions
- Keep separate memory per user
- Fetch a stored photo by description
Pricing
Freemium, from $5 a month
Free with $5 credit after a survey; Mini $5/mo, Pro $20/mo, Max from $100/mo
Prices change. Check wontopos.com before you buy.
MCP FAQ
Is it beta?
The registry description labels it beta.
What do I need to run it?
A Wontopos API key set in WONTOPOS_API_KEY.
Is there a free tier?
Free gives $5 credit after a short survey, then pay as you go.
Is my data used for training?
The pricing page says memories are never used to train AI models.
Is MCP free?
MCP has a free plan, with paid plans for more usage or features. Paid plans start at $5 a month.
How do I add MCP to Claude Code?
Run this command in your terminal: claude mcp add mcp --env WONTOPOS_API_KEY='<WONTOPOS_API_KEY>' --env WONTOPOS_HTTP_TOKEN='<WONTOPOS_HTTP_TOKEN>' --env WONTOPOS_ROUTER_KEY='<WONTOPOS_ROUTER_KEY>' --env WONTOPOS_SKILL_KEY='<WONTOPOS_SKILL_KEY>' -- npx -y wontopos-mcp
What are the best alternatives to MCP?
The closest alternatives to MCP are Codebase Memory, XERJ, Claude Code Ultimate Guide, Nautilus Compass Memory and note1.
Alternatives to MCP

Codebase Memory
MCP serverLocal MCP server that builds a code knowledge graph for AI coding agents
Runs locally- Open source
- 48.7K GitHub stars
- Knowledge and memory

XERJ
MCP serverLocal search engine for AI agents with cited answers over your code and docs
Runs locally- Open source
- 19.2K GitHub stars
- Knowledge and memory

Claude Code Ultimate Guide
MCP serverMCP server for searching the Claude Code Ultimate Guide from any MCP client
Runs locally- Open source
- 6.5K GitHub stars
- AI models and LLM tooling

Nautilus Compass Memory
MCP serverLocal-first long-term memory layer for AI agents, served over MCP
Not checked yet- 5 tools
- Remote
- Open source
- 1.3K GitHub stars
- Knowledge and memory

note1
MCP serverMCP server giving AI assistants search and summaries across your note1 meetings
- 4 tools
- Remote
- Free trial
- 1 GitHub star
- Productivity
MedXpert 图书馆连接器(通用版)
MCP serverLocal MCP server that searches and answers questions over your own Markdown wiki
Runs locally- 4 tools
- Open source
- Knowledge and memory