Practical overview
When it fits
Use Marm Memory when an MCP-compatible coding assistant needs searchable project history or indexed repository context beyond a single chat. Choose HTTP when multiple agents should share a server, or STDIO for a local single-agent process.
Who it is for
- Developers who want coding assistants to retrieve decisions, notes and project history across sessions.
- Teams or individual builders connecting multiple assistants to shared code and memory context.
- People who want to index a repository so an MCP-connected assistant can find symbols and follow code paths.
Documented uses
- Store conversations, notes, notebook entries and summaries for later search.
- Index a code repository and expose its symbols and code paths to assistants.
- Connect stored concepts such as decisions, errors and ideas, with links to relevant code where available.
- Share one memory service across multiple agents through HTTP.
Requirements & limits
The repository documents pip installation, Docker options, and local HTTP and STDIO modes. It is licensed under Apache-2.0.
- HTTP and STDIO expose the same 16 tools, but HTTP supports a shared server while STDIO is a local single-process setup.
- The documented local pip HTTP server binds to localhost by default and does not require a key; exposing it beyond localhost requires a key.
- Docker HTTP is documented for shared or always-on use and requires a key.
- The project is written in Python and stores its combined memory and graph data in SQLite.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up local MARM memory over HTTP
Install MARM Memory, start its local HTTP server, and connect an MCP client.
Before you start
- Python 3.10 or later, as indicated by the project README.
- An MCP client with a command-line interface; Codex uses a separate connection command.
Steps
Install the package with the command below.
pip install marm-mcp-serverStart the local server with
marm-memory start.For most MCP clients, connect using
"agent" mcp add --transport http marm-memory http://localhost:8001/mcp, replacingagentwith your client's CLI command.For Codex, connect using
codex mcp add marm-memory --url http://localhost:8001/mcp.
Expected outcome
The MCP client is configured to connect to the local MARM HTTP server.
Guide limitations
- The documented default binds to localhost and does not require a key unless you expose the server with
SERVER_HOST=0.0.0.0. - The source describes this as a local single-machine setup; it does not document client-specific commands beyond the examples.
Set up local MARM memory over STDIO
Install MARM Memory and configure an MCP client to use its local STDIO server.
Before you start
- Python 3.10 or later, as indicated by the project README.
- An MCP client that supports STDIO; the README lists Claude Code, Cursor, VS Code, Qwen, and Antigravity.
Steps
Install the package with the command below.
pip install marm-mcp-serverStart the STDIO server with the command below.
python -m marm_mcp_server.server_stdioFor most MCP clients, configure the connection with
"agent" mcp add --transport stdio marm-memory-stdio marm-mcp-stdio.For Codex, use
codex mcp add marm-memory-stdio -- marm-mcp-stdio; when using a virtual environment or path-based setup, the README says to replacemarm-mcp-stdiowith the command below.python -m marm_mcp_server.server_stdio
Expected outcome
The MCP client is configured for a local STDIO process exposing MARM's tools.
Guide limitations
- STDIO is documented as a single local process for private, single-agent use; it uses no port and requires no API key.
- The README notes that virtual-environment or path-based setups may need the Python module command instead of
marm-mcp-stdio.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- Lyellr88/marm-memory · Official homepage title · retained in its source language