huml

    1

    Persistent cross-session memory for AI agents using the huml MCP server.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    huml

    Example session with this skill installed

    Set up huml so Claude Code and Cursor share one memory.

    • Read your context and instructions
    • Compiled the huml

    huml is deployed to your Cloudflare account and connected to both Claude Code and Cursor. They now read and write the same memory.

    What I did

    StepCommand / changeResult
    Clone and installgit clone https://github.com/CodeBlastr/huml && cd huml && npm installDone
    Env fileCopied .env.example to .envRequired by the deploy scripts
    Cloudflare authnpx wrangler loginYou approved in the browser
    Databasenpx wrangler d1 create humlID copied into wrangler.toml
    OAuth storenpx wrangler kv namespace create OAUTH_KVID copied into wrangler.toml
    URLworkers_dev = true, routes removed, PUBLIC_URL sethttps://huml.<your-subdomain>.workers.dev
    Tables and deploynpm run migrate:remote then npm run deployLive
    Root tokenopenssl rand -hex 32 | npx wrangler secret put AUTH_TOKENSet, never displayed
    Client tokensYou ran npm run tokens -- issue claude-code, issue cursor and issue scriptsTokens stayed out of this session

    Client config

    You pasted your tokens into these yourself.

    Claude Code

    claude mcp add --transport http --scope user huml https://huml.<your-subdomain>.workers.dev/mcp \
      --header "Authorization: Bearer <claude-code token>"
    

    Cursor (~/.cursor/mcp.json):

    {
      "mcpServers": {
        "huml": {
          "url": "https://huml.<your-subdomain>.workers.dev/mcp",
          "headers": { "Authorization": "Bearer <cursor token>" }
        }
      }
    }
    

    Verified

    • npm run smoke passed: created, read and deleted a test doc under /_smoke/
    • Wrote /preferences/setup.md from Claude Code and read it back from Cursor

    Next steps

    • Restart Cursor so it picks up the new MCP server
    • Import existing notes with npm run seed -- ./notes
    • Add this line to CLAUDE.md and Cursor rules: "Check huml memory for context about people, projects and my preferences before answering, and save anything worth remembering."
    • Revoke a client anytime with npm run tokens -- revoke <id>

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Standard LLM sessions are ephemeral, losing context the moment you close the tab. This skill provides a persistent memory layer for your AI agents using the huml MCP protocol, allowing them to store and retrieve long-term data across different sessions, IDEs, and chat interfaces.

    What it does

    • Persistent Storage manages long-term facts, project statuses, and user preferences in Markdown files.
    • Cross-Client Sync enables shared context between Claude, ChatGPT, Cursor, and CLI tools via a central MCP server.
    • Conflict Resolution uses compare-and-swap logic to prevent data loss during concurrent writes from multiple agents.
    • Lexical Search provides full-text search capabilities to find specific documents by name, alias, or content snippets.
    • Context Management automatically reads relevant project history or person-specific notes before starting new tasks.

    How it works

    1. Discovery involves using memory_list or memory_search to find existing records related to the current task.
    2. Retrieval fetches the full content and version ID of relevant files using memory_read.
    3. Modification updates data using memory_append for logs or memory_str_replace for specific edits, passing the version ID to ensure consistency.
    4. Creation initializes new memory entries with YAML frontmatter using memory_write and a "new" version flag.

    Frameworks & tools

    This skill requires a running huml server acting as an MCP (Model Context Protocol) host. It interacts with the server via standard MCP tool calls and manages Markdown files with YAML frontmatter stored in a SQLite-backed environment.

    Why this beats prompting it yourself

    Manual prompting cannot bridge the gap between separate chat sessions or different AI tools. This skill provides the logic for safe concurrent writes and structured data management that prevents agents from hallucinating past interactions.

    Use cases

    • Tracking project status and blockers across multiple developer tools.
    • Storing user-specific coding standards and architectural preferences.
    • Maintaining dossiers on clients, teammates, or stakeholders.
    • Logging architectural decisions and their underlying rationales.

    Known limitations

    Requires an active huml MCP server deployment. Does not support semantic search, only lexical word matching. Files are capped at 100KB each.

    Requires an external huml MCP server. Lexical search only; no semantic/vector retrieval. Files limited to 100KB. Does not store secrets or API keys.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    GitHub
    wrangler
    Node.js
    Git

    Permissions

    Allowed Hosts

    github.com

    File Scopes

    __MACOSX/**
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Huml solves the problem of "agent amnesia" by providing a persistent Markdown-based storage layer that allows AI agents to remember project status, user preferences, and past decisions across different sessions and chat interfaces.

    This skill is compatible with any AI client or IDE that supports the Model Context Protocol (MCP), including Claude Desktop, Cursor, and various CLI tools, enabling a unified memory across all your development environments.

    Huml uses a sophisticated compare-and-swap logic with version IDs to manage concurrent writes, ensuring that data is never overwritten or lost even if multiple agents are updating the same memory file simultaneously.

    Setup requires a running huml MCP server; once active, you simply connect the skill to your host, allowing the agent to immediately begin using the memory_list, search, and write tools to manage your data.

    Huml currently utilizes lexical search for precise keyword and alias matching within your Markdown files, though it does not yet support vector-based semantic search.

    Memory files are stored as individual Markdown documents with YAML frontmatter, with a size limit of 100KB per file to ensure optimal performance and context window management for your agents.

    New in LLM Ops