memora-mcp

MCP server for saving and retrieving prompts, memories, and plans. Gives your AI agent a persistent knowledge repository that survives across sessions.

Installation

npx memora-mcp

Tools (17)

Memories — facts, decisions, and context

| Tool | Description | | ------------------ | --------------------------------------------------------------------------------------------- | | save_memory | Persist a fact, decision, observation, or context that should survive across sessions | | get_memory_by_id | Fetch the full content of a memory by its exact ID | | list_memories | List saved memories (id, title, summary, tags — content omitted). Supports tag filter + pagination | | update_memory | Partially update a memory — omitted fields are preserved | | delete_memory | Permanently delete a memory |

Prompts — reusable instructions and templates

| Tool | Description | | ------------------ | --------------------------------------------------------------------------------------------- | | save_prompt | Save a reusable instruction, system prompt, or template | | get_prompt_by_id | Fetch the full text of a prompt by its exact ID | | list_prompts | List saved prompts (id, title, summary, tags — content omitted). Supports tag filter + pagination | | update_prompt | Partially update a prompt — bump version when making breaking changes | | delete_prompt | Permanently delete a prompt |

Plans — multi-step implementation plans and ADRs

| Tool | Description | | --------------- | ------------------------------------------------------------------------------------------ | | save_plan | Save a structured, multi-step implementation plan or architectural decision record | | get_plan_by_id| Fetch the full content of a plan by its exact ID | | list_plans | List saved plans (id, title, summary, tags — content omitted). Supports tag filter + pagination | | update_plan | Partially update a plan — use to mark steps complete or revise scope | | delete_plan | Permanently delete a plan |

Global search

| Tool | Description | | --------------------- | ------------------------------------------------------------------------------------------------- | | search_knowledge | Free-text search across all items (memories, prompts, plans). Matches title, summary, content, and tags | | get_recent_knowledge| Retrieve the most recently updated items across all types — useful at session start to regain context |

Configuration

Claude Desktop

Add to your claude_desktop_config.json:

With external API (recommended for production)

{
  "mcpServers": {
    "memora": {
      "command": "npx",
      "args": ["-y", "memora-mcp"],
      "env": {
        "STORAGE_TYPE": "http",
        "SAVE_PROMPT_API_URL": "https://your-api.com",
        "API_KEY": "your-api-key-here"
      }
    }
  }
}

With local JSON file (no API required)

{
  "mcpServers": {
    "memora": {
      "command": "npx",
      "args": ["-y", "memora-mcp"],
      "env": {
        "STORAGE_TYPE": "json"
      }
    }
  }
}

Data is stored at ~/.save-prompt/data.json.

With local Markdown files (human-readable, git-friendly)

{
  "mcpServers": {
    "memora": {
      "command": "npx",
      "args": ["-y", "memora-mcp"],
      "env": {
        "STORAGE_TYPE": "markdown"
      }
    }
  }
}

Each item is stored as an individual .md file at ~/.save-prompt/items/<uuid>.md with YAML frontmatter for metadata and a markdown body for content. Files are human-readable and can be edited directly or committed to git.

VS Code / Cursor

Add to your .vscode/mcp.json (or user settings):

{
  "servers": {
    "memora": {
      "command": "npx",
      "args": ["-y", "memora-mcp"],
      "env": {
        "STORAGE_TYPE": "json"
      }
    }
  }
}

Environment Variables

| Variable | Required | Default | Description | | --------------------- | ---------------------- | ------- | ----------------------------------- | | STORAGE_TYPE | No | http | http, json, or markdown | | SAVE_PROMPT_API_URL | If STORAGE_TYPE=http | — | API base URL | | API_KEY | If STORAGE_TYPE=http | — | Sent as x-api-key header | | API_TIMEOUT | No | 10000 | HTTP request timeout in milliseconds |

API Contract

When using STORAGE_TYPE=http, your API must implement:

| Method | Path | Description | | -------- | ----------------------- | ---------------------------------------- | | POST | /knowledge-items | Create item | | GET | /knowledge-items/:id | Get by ID (404 if not found) | | GET | /knowledge-items | List with filters (see query params below) | | PATCH | /knowledge-items/:id | Partial update | | DELETE | /knowledge-items/:id | Delete |

List query parameters:

| Parameter | Type | Description | | --------- | ------ | -------------------------------------------- | | type | string | Filter by prompt, memory, or plan | | tags | string | Comma-separated tags (ALL must match) | | query | string | Free-text search across title, summary, content, and tags | | limit | number | Max items to return (1–100) | | offset | number | Items to skip for pagination | | orderBy | string | createdAt, updatedAt, or title |

KnowledgeItem schema:

{
  "id": "uuid",
  "type": "prompt | memory | plan",
  "title": "string",
  "summary": "string (max 500 chars, optional) — shown in list views",
  "content": "string (markdown, JSON, or plain text)",
  "tags": ["string"],
  "version": "string",
  "createdAt": "ISO8601",
  "updatedAt": "ISO8601"
}

List and search responses omit content for context efficiency. Use get_*_by_id to fetch the full content of a specific item.

Development

npm install
npm run build
npm test

License

MIT