Compare
total-agent-memory vs mcp-memory-service
mcp-memory-service is a self-hosted memory backend with REST, MCP, OAuth and a knowledge graph, aimed at LangGraph, CrewAI and AutoGen pipelines as well as Claude. total-agent-memory is an MCP server for the coding agents on your laptop and a team server for your company. Where they overlap and where they do not.
Facts about mcp-memory-service were checked on October 9, 2026 against its public source. Facts about total-agent-memory are current for the version in the site header.
The short version
mcp-memory-service is a self-hosted memory backend that speaks REST, MCP, OAuth and a CLI, with a dashboard and a knowledge graph. It is built for agent pipelines (LangGraph, CrewAI, AutoGen, any HTTP client) and also connects to Claude Desktop, Claude Code and OpenCode. Embeddings run locally through ONNX. Storage can be SQLite, Cloudflare, or a hybrid of the two with background sync. It also supports remote MCP over OAuth so claude.ai and ChatGPT in the browser can use it.
total-agent-memory (TAM) is narrower on purpose. It is an MCP server for the coding agents on your machine, Claude Code, Codex CLI, Cursor and six others, with one SQLite file, local embeddings and no LLM on the save or search path. For companies there is a separate team server with personal, team and shared scopes.
Side by side
| mcp-memory-service | total-agent-memory | |
|---|---|---|
| Built for | Agent pipelines and Claude clients | Coding agents, then a team of them |
| Transports | REST API, MCP, OAuth remote MCP, CLI | MCP (stdio and the team server’s /mcp), CLI search |
| Install | pip install mcp-memory-service, then configure the client | One command per client wires it: npx -y total-agent-memory connect cursor |
| Storage | SQLite, Cloudflare, or hybrid with cloud sync | One SQLite file; team server with PostgreSQL or SQLite |
| Embeddings | Local ONNX | Local MiniLM, downloaded once |
| Knowledge graph | Typed edges such as causes, fixes, contradicts | Entities, relations, facts with validity dates |
| LLM in the loop | Optional quality scoring via an OpenAI-compatible endpoint | Optional enrichment, never on the save or search path |
| Browser clients | claude.ai and ChatGPT over remote MCP | Not a target |
| Licence | Apache-2.0 | MIT |
| Community | 2,006 GitHub stars | 72 GitHub stars |
Facts about mcp-memory-service come from its README on the date above. If something is out of date, open an issue and this page gets corrected.
Benchmarks
mcp-memory-service does not publish a LongMemEval or LoCoMo result that we could find on the date above, so there is nothing to put next to ours. TAM’s numbers and their limits are on the benchmarks page: LongMemEval R@5 95.1% on version 13, per-question files public, and a LoCoMo QA run against Mem0 OSS that shows no established advantage.
When to pick mcp-memory-service
- Your agents are a pipeline, not an IDE assistant, and need a REST API.
- You want memory in the browser through claude.ai or ChatGPT with OAuth.
- You want cloud sync across devices through Cloudflare.
When to pick total-agent-memory
- You want a coding agent wired in one command and a single file you can back up or delete.
- You want the rules of what to save and when to recall shipped with the server: the
memory-protocolskill and capture hooks for Claude Code. - You need a team server with scopes, tokens and an audit trail that you run yourself.
Install takes one command: see the quick start.