MemoryLake Skills

A skill that enables AI agents to create projects, upload documents, and then search/retrieve/analyze data from a MemoryLake server — the memory layer for AI Agents.

Compatible Platforms

This skill works with any AI coding agent that supports MCP or can execute shell commands:

Claude Code	Cursor	OpenAI Codex	Manus	Google Antigravity	OpenClaw
Anthropic's agentic CLI for Claude	AI-powered code editor	OpenAI's terminal coding agent	Autonomous AI agent by Meta	Google's agent-first IDE with Gemini	Open-source personal AI assistant

How it works:

• Use the REST API (Bearer API key + X-User-ID) to create a project, upload documents, and generate an MCP secret.
• Use the Streamable HTTP MCP server (secret in query string) for search + retrieval + code-run analysis.

This repo includes two shell clients:

• scripts/memorylake_rest_client.sh (REST)
• scripts/memorylake_client.sh (MCP)

What is MemoryLake?

MemoryLake ingests unstructured files (Excel, PDF, text, etc.), chunks and indexes their content, and exposes it through a Streamable HTTP MCP Server for intelligent retrieval and data analysis. It gives AI agents long-term memory over your documents.

What This Skill Does

• Hybrid search — Semantic + keyword (BM25) search across all indexed documents
• File exploration — Browse memorylake contents, file types, and metadata
• Code-based analysis — Run Python (pandas, numpy, scikit-learn, etc.) against your data server-side
• Session management — A bash client that handles the MCP Streamable HTTP protocol (JSON-RPC, SSE, session lifecycle)

Project Structure

memorylake-skills/
├── SKILL.md                          # Skill definition with end-to-end workflow (REST → MCP)
├── scripts/
│   ├── memorylake_rest_client.sh      # REST client (projects, uploads, documents)
│   └── memorylake_client.sh           # MCP client (Streamable HTTP MCP)
├── references/
│   ├── mcp-tools.md                   # MCP tool parameters and response formats
│   └── memorylake-openapi.json         # REST OpenAPI spec (projects/drives/upload/etc.)
└── memorylake-skills.skill             # Packaged distributable (.zip with .skill extension)

Quick Start

1. Install the skill

Download memorylake-skills.skill from Releases or clone this repo, then install in your AI agent. For example, in Claude Code:

/skill install memorylake-skills.skill

2. Get a MemoryLake API key (REST)

1. Go to https://app.memorylake.ai/ and apply for an API key.
2. REST Base URL:

https://app.memorylake.ai/openapi/memorylake

Auth headers:

• Authorization: Bearer <your API key>
• X-User-ID: <your user id>

3. Create a project + generate MCP secret

export MEMORYLAKE_BASE_URL="https://app.memorylake.ai/openapi/memorylake"
export MEMORYLAKE_API_KEY="<your api key>"
export MEMORYLAKE_USER_ID="<your user id>"

# Create project
./scripts/memorylake_rest_client.sh projects:create '{"name":"My Project"}'

# Create project API key (this is the MCP secret)
./scripts/memorylake_rest_client.sh projects:create-apikey <project_id> '{"description":"mcp"}'

Use the returned secret to form the MCP URL:

https://ai.data.cloud/memorylake/mcp/v1?apikey=<secret>

4. Upload documents + quick-add into project

# Create multipart upload and complete it (uploads parts to presigned URLs)
./scripts/memorylake_rest_client.sh upload:create-multipart '{"file_size": 123456}' > upload.json
./scripts/memorylake_rest_client.sh upload:complete-multipart upload.json /path/to/file.pdf

# Then quick-add into the project
./scripts/memorylake_rest_client.sh projects:quick-add <project_id> '{"object_key":"<object_key>","file_name":"file.pdf"}'

If you have multiple documents, upload + quick-add one by one.

5. Poll document status until ready

./scripts/memorylake_rest_client.sh projects:list-documents <project_id>

Document status values: error, okay, running, pending. Recommended polling interval: 5s.

6. Ask questions about your data (MCP)

Once installed, your AI agent can:

• "What files are in my memorylake?"
• "Search for budget reports from 2024"
• "Analyze the revenue data across all uploaded spreadsheets"
• "Summarize the content of my PDF documents"

Available Tools

The skill exposes 5 MemoryLake tools via HTTP:

Tool	Purpose
get_memorylake_metadata	Explore memorylake structure and contents
search_memory	Hybrid semantic + keyword search across all files
fetch_memory	Get detailed metadata for specific files
create_memory_code_runner	Create a Python execution environment
run_memory_code	Run Python code against your data (pandas, numpy, etc.)

Using the MCP Client Script Standalone

Once you have the MCP URL (.../mcp/v1?apikey=<secret>), you can use scripts/memorylake_client.sh directly:

MCP_URL="https://ai.data.cloud/memorylake/mcp/v1?apikey=<secret>"

# Initialize a session
SESSION=$(./scripts/memorylake_client.sh "$MCP_URL" init)

# Explore your memorylake
./scripts/memorylake_client.sh "$MCP_URL" "$SESSION" get_memorylake_metadata

# Search for content
./scripts/memorylake_client.sh "$MCP_URL" "$SESSION" search_memory '{
  "parsed_query": {
    "bm25_cleaned_query": "revenue report 2024",
    "named_entities": [],
    "bm25_keywords": ["revenue", "report", "2024"],
    "bm25_boost_keywords": ["revenue", "2024"],
    "rewritten_query_for_dense_model": "Annual revenue and financial reports for the year 2024"
  }
}'

# Fetch file details
./scripts/memorylake_client.sh "$MCP_URL" "$SESSION" fetch_memory '{"memory_ids": ["ds-abc123"]}'

Requirements

• curl (included on macOS and most Linux distros)
• A MemoryLake API key from memorylake.ai

Links

• MemoryLake — The memory layer for AI Agents
• Model Context Protocol — The protocol MemoryLake implements
• Claude Code | Cursor | OpenAI Codex | Manus | Google Antigravity | OpenClaw

License

MIT