# mcp-memory-service

> Open-source persistent memory for AI agent pipelines (LangGraph, CrewAI, AutoGen) and Claude. REST API + knowledge graph + autonomous consolidation.

- **URL**: https://www.freshcrate.ai/projects/mcp-memory-service
- **Author**: doobidoo
- **Category**: MCP Servers
- **Latest version**: `v10.65.0` (2026-05-24)
- **License**: Apache-2.0
- **Source**: https://github.com/doobidoo/mcp-memory-service
- **Language**: Python
- **GitHub**: 1,712 stars, 259 forks
- **Registry**: github
- **Tags**: `agent-memory`, `agentic-ai`, `ai-agents`, `autogen`, `claude`, `crewai`, `knowledge-graph`, `langgraph`, `python`

## Description

Open-source persistent memory for AI agent pipelines (LangGraph, CrewAI, AutoGen) and Claude. REST API + knowledge graph + autonomous consolidation.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v10.65.0` | 2026-05-24 | High | ## What's New  OpenCode plugin gains three slash commands, a working TUI sidebar widget, and a per-instance status bridge.  ### Added  - **`/memory` slash commands** — three commands via `command.execute.before` hook:   - `/memory` — current session status (memories loaded, last capture, server health)   - `/memory search <query>` — top 5 semantic matches from the vector DB   - `/memory health` — backend type, status, total memory count (hits `/api/health/detailed`) - **TUI sidebar widget** (`op |
| `v10.59.2` | 2026-05-17 | High | ## What's Changed  ### Bug Fixes  **fix(oauth): use AnyUrl for redirect_uri in AuthorizationRequest and TokenRequest** (#942, reported by @tkislan)  `HttpUrl` only accepts `http`/`https` schemes — `cursor://`, `vscode://`, `vscode-insiders://` were silently rejected by Pydantic before reaching the `ALLOWED_SCHEMES` whitelist in `registration.py`. This meant the IDE scheme support added in v10.59.0 was a no-op in practice: Pydantic discarded the URI before the application code ever saw it.  **Fix |
| `v10.54.0` | 2026-05-10 | High | ## What's New  ### feat(search): `tag_match` parameter for `memory_search` MCP tool  Extends the AND/OR tag filtering already available in `memory_delete` to the `memory_search` MCP tool.  **Usage:** - `tag_match: "any"` (default, OR) — return memories matching at least one of the supplied tags. Preserves existing behavior, no breaking change. - `tag_match: "all"` (AND) — return only memories that have every supplied tag.  **Example:** ```json {   "query": "python development",   "tags": ["pytho |
| `v10.49.4` | 2026-05-05 | High | ## Summary  Consolidation safety fix: high-value mistake notes no longer vanish during scheduled consolidation passes.  ## What's Changed  ### Fixed  - **[#853] Mistake-notes survive consolidation**: `_is_protected_memory()` in `consolidation/base.py` now shields memories with `memory_type='mistake'` and `failure_count >= 3` from decay and forgetting passes. Error-replay records that have been marked as significant failures are never eligible for archival or removal by the consolidation schedule |
| `v10.45.0` | 2026-04-30 | High | ## What's New  ### feat(quality): OpenAI-compatible quality scoring provider (PR #790)  Point quality scoring at any OpenAI `/v1/chat/completions`-compatible endpoint — Ollama, LiteLLM, MLX-LM server, or vLLM — without a cloud API key or the ONNX model.  **Quick start:** ```bash MCP_QUALITY_AI_PROVIDER=openai-compatible MCP_QUALITY_AI_BASE_URL=http://localhost:11434/v1   # Ollama MCP_QUALITY_AI_MODEL=qwen2.5:7b-instruct # MCP_QUALITY_AI_API_KEY=ollama                     # optional ```  Recommen |
| `v10.40.3` | 2026-04-24 | High | ## What's Changed  ### Fixed  - **claude-hooks: socket hang-up on multi-phase retrieval eliminated** (`memory-client.js`): Node.js HTTPS agent defaults to `keepAlive: true`, which causes Uvicorn to close idle sockets after ~5 s. The hook is a one-shot CLI process — keepAlive provides zero benefit and caused subsequent phase requests to reuse dead sockets, producing `ECONNRESET` ("socket hang up"). Added `agent: false` and `Connection: close` header to `_attemptHealthCheck`, `storeMemoryHTTP`, an |
| `v10.40.0` | 2026-04-22 | High | ## Special Thanks  Huge thanks to @zc277584121 for contributing the Milvus storage backend — the first major community-contributed storage backend in the project's history. This is ~1,750 lines of production-quality code with 39 tests, three deployment modes, and a 6-month maintenance SLA commitment. This is exactly the kind of deep, thoughtful contribution that expands the project's reach to new deployments.  ## What's New  ### Added  - **[#721] Milvus storage backend (Lite / self-hosted / Zill |
| `v10.39.1` | 2026-04-20 | High | ## Hotfix for v10.39.0 plugin install  **If you tried `/plugin install mcp-memory-service` on v10.39.0 and hit `Validation errors: author: Invalid input: expected object, received string`, this release fixes it.** The `plugin.json` manifest now uses the object form (`{"name": "..."}`) the Claude Code plugin validator requires.  Thanks to @yingzhi0808 for the detailed bug report (#738) and the one-line fix (#739).  ## What Changed  ### Fixed  - **plugin**: `plugin.json` `author` field now uses th |
| `v10.39.0` | 2026-04-19 | High | ## New: Install as a Claude Code Plugin  Two commands replace the bash wrapper, API-key generation, and manual MCP-config-editing ceremony:  \`\`\` /plugin marketplace add doobidoo/mcp-memory-service /plugin install mcp-memory-service \`\`\`  v1.0.0 plugin is experimental. See [`claude-hooks/PLUGIN.md`](https://github.com/doobidoo/mcp-memory-service/blob/main/claude-hooks/PLUGIN.md) for install, config, and migration from the legacy `install_hooks.py` installer. Closes #530.  ## What's Changed |
| `v10.38.4` | 2026-04-19 | High | ## fix(mcp): return HTTP 202 for JSON-RPC notifications — strict-client compatibility  ### Problem  The `/mcp` Streamable HTTP endpoint was returning `-32601 Method not found` in response to JSON-RPC **notifications** (messages without an `id` field), specifically `notifications/initialized`. This violated two specs:  - **JSON-RPC 2.0 §4.1**: servers MUST NOT reply to notifications - **MCP Streamable HTTP spec**: notifications must receive `HTTP 202 Accepted` with an empty body  Claude Code and |

## Citation

- HTML: https://www.freshcrate.ai/projects/mcp-memory-service
- Markdown: https://www.freshcrate.ai/projects/mcp-memory-service.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/mcp-memory-service/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._