Memanto MCP
The Memanto memory MCP server: an agent stores facts and decisions and finds them in later sessions, even as of a date.
- 2,363 ★ on the repository
- Last commit
Extended access
Why this level
- Stored entries are sent to the Moorcheh cloud service and kept there.
- Over HTTP or SSE the server does not authenticate incoming connections.
Install
- Works in
Manual install
pip install memanto-mcpRequires a Moorcheh key in the `MOORCHEH_API_KEY` variable.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Stored entries are sent to the Moorcheh cloud service and kept there.
- Over HTTP or SSE the server does not authenticate incoming connections.
This is third-party code. Review the repository files before installing.
What it does
The `memanto-mcp` package exposes Memanto memory functions as MCP tools. Seven are available by default: `remember` stores one entry (a fact, decision, preference, goal or instruction), `batch_remember` takes up to a hundred at once, `recall` searches by meaning, `recall_recent` shows the latest, `recall_as_of` restores knowledge at a date, `recall_changed_since` returns changes since the last check and `answer` synthesizes an answer from memory. Memory is bound to an agent. The first call creates the agent from `MEMANTO_DEFAULT_AGENT_ID` and opens a self-renewing session. Each entry has a source: when a call omits it, the server fills in the name of the connected client, so Cursor and Codex stay distinguishable in recall. Four agent-management tools are enabled with `MEMANTO_EXPOSE_ADMIN=true`.
Who it is for. People who work with several agents or editors and want project decisions, preferences and agreements not to be lost between sessions.
Good fit when
- An agent records a decision to move to JWT, and another finds it by query a week later.
- Before starting work you need to recall the user's stable preferences for answer format.
- A team checks what agents learned about a project since the last release.
- Dozens of facts are extracted from a document and stored in one batch call.
Not a fit when
- Storing entries in the Moorcheh cloud is not acceptable and fully local memory is required.
- Agents only need the memory of a single conversation.
- The MCP server would be exposed publicly with no reverse proxy and authorization.
Example request
Remember that we moved to JWT and dropped session cookies, and remind me tomorrow why we decided so.MCP
- Transport
- stdio, http, sse
- Authentication
- API key
Environment variables
MOORCHEH_API_KEYrequired, secret- Moorcheh API key.
MEMANTO_DEFAULT_AGENT_ID- Default agent so calls can omit `agent_id`.
MEMANTO_EXPOSE_ADMIN- Enables the four agent-management tools.
Limitations
It needs a Moorcheh key (the free tier gives 100 thousand operations a month), Python 3.10 or newer and the `memanto` package at 0.2.13 or later. Writes and search go through the Moorcheh cloud service over HTTPS, so stored facts leave the machine. The server does not authenticate inbound clients: over HTTP or SSE it needs a reverse proxy with authorization. The first call is slower because of the SDK import and session setup.
How to disable. Remove the memanto entry from the MCP configuration and run `pip uninstall memanto-mcp`. Stored entries stay in the Moorcheh service, so delete them through the Memanto CLI or remove the agent with the `delete_agent` tool.
FAQ
How do I connect Memanto MCP to Cursor?
Install `pip install memanto-mcp` and add a `memanto` server to `~/.cursor/mcp.json` with the command `memanto-mcp` and the variables `MOORCHEH_API_KEY` and `MEMANTO_DEFAULT_AGENT_ID`. The tools appear after a restart.
How do `recall_as_of` and `recall_changed_since` differ?
`recall_as_of` restores what was known at a given date, for example 1 November 2025. `recall_changed_since` shows only what is new since the last check.
How do I run the server over HTTP?
Run `memanto-mcp --transport streamable-http --host 0.0.0.0 --port 8765`. For SSE switch the transport to `sse`. Put the port behind a reverse proxy with authorization, since the server does not check clients.
How does it differ from the Memanto CLI?
The CLI runs the full memory agent and the `memanto connect claude-code` command, while `memanto-mcp` exposes memory functions as tools only. Both use the same SDK client.
What is the entry source?
The `source` field names who wrote an entry: `user`, `agent`, `tool`, `system` or a client such as `cursor` or `codex`. Recall can be filtered by it. When a call omits the source, the server uses the connected client name.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for long-term agent memory: compresses dialogues into atomic facts, resolves pronouns and anchors relative time to absolute dates
Long-term memory for Claude Code, Cursor and Codex: the agent saves decisions and finds them again in later sessions
A long-term memory MCP server for agents: search, store and correct facts across sessions in the cloud or a local core
A self-hosted MCP server for long-term agent memory: structured records, versioned rollback and a visual dashboard
Engineering team memory over MCP: decisions, bug causes and constraints recalled by every agent, from Claude Code to Cursor.
An MCP server for a project memory bank: the agent stores and reads memory files through one service instead of inside the project
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Memanto MCP (this card) | yes | API key | 2,363 | |
| Supermemory | yes | OAuth | 31,595 | a remote memory server with plugins and an API; Memanto MCP adds as-of and changed-since recall and needs a Moorcheh key for the cloud |
| Basic Memory | no | not required | 4,557 | keeps memory in local Markdown files; Memanto MCP holds entries in a cloud service with types and sources |
README in short
The server lives in the `integrations/mcp` folder of the Memanto repository and uses the same SDK client as the CLI. Configuration is by environment variables or a `.env` file.