# Loom > Makes AI stateful across sessions. Workflow: chat → idea → design → reqs → plan → do_step(s) → done, with fresh scoped context, staleness detection, and human approval at every phase. Roadmap & history of shipped features. - **Type:** MCP server - **Install:** `agentstack add mcp-reslava-loom` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [reslava](https://agentstack.voostack.com/s/reslava) - **Installs:** 0 - **Category:** [Developer Tools](https://agentstack.voostack.com/c/developer-tools) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [reslava](https://github.com/reslava) - **Source:** https://github.com/reslava/loom ## Install ```sh agentstack add mcp-reslava-loom ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Loom **Your AI coding agent forgets everything every session. Loom is the memory it doesn't have.** Docs are the database. Your AI reads them, plans in them, and picks up exactly where it left off — across a fresh session, with your project's full context. No re-explaining. 🔗 **Get Loom:** [VS Code Marketplace](https://marketplace.visualstudio.com/items?itemName=reslava.loom-vscode) · [Open VSX](https://open-vsx.org/extension/reslava/loom-vscode) · [CLI on npm](https://www.npmjs.com/package/@reslava/loom) · [GitHub](https://github.com/reslava/loom) 📚 **Guides:** [Which way is for you?](./docs/WAYS-TO-USE-LOOM.md) · [Core concepts](./docs/USER_GUIDE.md) · [VS Code Extension](./docs/EXTENSION_USER_GUIDE.md) · [CLI / Claude Code](./docs/CLI_USER_GUIDE.md) --- ## The problem Every AI coding tool has the same flaw: **context is a growing garbage bag.** One long session piles up old decisions, dead ends, and half-finished threads — and the model degrades as it reasons over all of it at once. - **Session 1 is the best session.** By session 10 the AI has forgotten sessions 2–9. - **Re-explaining context every morning is expensive** — and letting the AI contradict a decision you made last week is worse. - There's no structure beneath the chat — just a window that grows forever with no memory of what was decided, or why. The cause isn't model quality. It's that there's no workflow underneath. ## What Loom does Loom replaces the chat window with a **document graph that _is_ the workflow.** Every idea, design, plan, and done-note is a typed, linked Markdown file. The AI reads exactly the right slice for the task in front of it — nothing more, nothing less. - **A session is a document graph, not a chat log.** Ideas → designs → plans with steps, all in Markdown, versioned in git. State is derived from the files; nothing lives only in the model's head. - **The AI resumes mid-plan.** Close the terminal, come back tomorrow, ask for the next step — it rereads the plan and continues with full context. No re-explaining what you're building. - **Specs propagate.** Change a design and its plans are flagged stale; ask the AI to *refine* and the change flows downstream. Context can't silently drift. - **Works with the agent you already use.** A VS Code extension for the visual flow — and it runs headless as a CLI and an MCP server for Claude Code, Cursor, or any MCP-capable agent. ## See it Threads + Context — your project as one graph: weave → thread → doc, each node carrying its derived state (locked reqs, step counts, staleness ⚠). The Context panel shows the exact docs — and their token cost — the AI will receive for the selected node, before you launch anything. Roadmap — the same graph re-laid as a derived roadmap: present + future threads in dependency-then-priority order, over a history of shipped versions. No list to maintain — it's computed from the documents. Click either image to view full size. ## Quick start **In VS Code (recommended — 1 click, no CLI, no Node):** 1. Install **Loom AI** from the [Marketplace](https://marketplace.visualstudio.com/items?itemName=reslava.loom-vscode) and click **Initialize Loom**. 2. Create a thread and a chat, and type what you want to build. 3. Click **AI Reply** to think it through, then **Generate Design** → **Generate Plan**. 4. Click **Do Step** — the AI implements the next step, writes a done-note, and marks it ✅. **From the terminal — a persisted Claude Code session + Loom "slang":** ```bash npx @reslava/loom install # scaffold loom/ + .mcp.json into your repo loom create thread ui dark-mode --title "Dark Mode" loom create chat ui dark-mode # a chat doc to think in ``` Now the distinctive part. Open the project in Claude Code and drive it with short **Loom slang** — each word maps to one action, and the AI writes every reply *back into the doc*, never into a terminal that scrolls away: - Describe the dark-mode feature in `chat-001.md` and ask for a design and plan, then say **`reply ui/dark-mode/chat-001.md`** → the AI answers *inside the chat* with the full thread loaded, and drafts the design + plan docs. - When the plan looks right, say **`do plan`** → the AI implements each step, marks it ✅, and writes a done-note — pausing for your `go` between steps. Close the terminal, come back tomorrow: the next session rereads the plan and picks up exactly where you stopped. → Full walkthrough: **[install to your first idea in five minutes](./loom/refs/getting-started-reference.md)**. ## Reports — query the *why*, not just the code Any tool with an LLM can summarize a codebase. What a codebase-only tool *can't* do is tell you **why** — which alternatives were weighed, why one won, where the code drifted from the design — because that reasoning was never in the code. It's in your chats, designs, and done-notes. Loom reads *those*. ```bash loom report project-overview # orient a newcomer, from the roadmap loom report decisions --weave auth # the "why" behind a weave's choices loom report release-notes # draft the next changelog, from the unreleased done plans ``` Each report is a versioned doc, and **every claim cites the source it came from**. → **[Reports reference](./loom/refs/reports-reference.md)**. ## Proven on real projects Loom is **built using Loom** — every feature in this repo went through its own loop (chat → idea → design → req → plan → done), living in the `loom/` graph next to the code it describes. It's also driving [**ChordFlow**](https://github.com/reslava/chord-flow), a separate, unrelated app whose entire music domain was designed in Loom chats before any code was written — the first real test that the workflow holds up outside this repo. ## Learn more | Guide | What's inside | |-------|---------------| | [Which way is for you?](./docs/WAYS-TO-USE-LOOM.md) | The four ways to run Loom (Guided · Power terminal · Pure agent · Automation) and how to pick | | [User Guide](./docs/USER_GUIDE.md) | Concepts, the workflow loop, and **how Loom decides what the AI sees** — the part that matters most | | [Extension User Guide](./docs/EXTENSION_USER_GUIDE.md) | The VS Code panel, buttons, and Context view | | [CLI / Claude Code Guide](./docs/CLI_USER_GUIDE.md) | Driving Loom from the terminal via an MCP agent | | [Architecture](./docs/ARCHITECTURE.md) · [AI Integration](./docs/AI_INTEGRATION.md) | How the layered system fits together, and how an agent plugs in via MCP | | [All reference docs](./loom/refs/) | The deep material: context pipeline, requirements model, staleness, MCP surface, command lists | > 💬 **Using Loom? I'd love to know how it's going.** Click **Feedback** in the status bar or run > `loom feedback` — it opens a prefilled GitHub issue you edit before sending. Opt-in, nothing sent > automatically. --- ## License MIT © 2026 Rafa Eslava ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [reslava](https://github.com/reslava) - **Source:** [reslava/loom](https://github.com/reslava/loom) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-reslava-loom - Seller: https://agentstack.voostack.com/s/reslava - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.