diff --git a/.github/release-notes-template.md b/.github/release-notes-template.md
new file mode 100644
index 0000000..8d27503
--- /dev/null
+++ b/.github/release-notes-template.md
@@ -0,0 +1,24 @@
+# Bindery {{VERSION}}
+
+## Highlights
+
+- Standardized release packaging across all supported surfaces.
+- See attached assets below for install-ready downloads.
+
+## Release Assets
+
+- bindery-{{VERSION}}.mcpb (Claude Desktop / Cowork MCP package)
+- bindery-mcp-server-{{VERSION}}.zip (Standalone MCP package for clients that support command + args + env configuration, such as ChatGPT Work and LM Studio)
+- bindery-{{VERSION}}.vsix (VS Code extension)
+- bindery-obsidian-bundled-{{VERSION}}.zip (Obsidian plugin bundle)
+- obsidian-plugin/dist/main.js and obsidian-plugin/dist/manifest.json (raw Obsidian artifacts for obsidian automated scanning)
+
+## Installation
+
+- VS Code extension and MCP setup overview: https://github.com/evdboom/Bindery/blob/main/README.md
+- MCP package details and manual install: https://github.com/evdboom/Bindery/blob/main/mcpb/README.md
+- Standalone MCP zip quick install: included README.md inside bindery-mcp-server-{{VERSION}}.zip
+
+## Notes
+
+- Release notes below this section are auto-generated from merged pull requests and commits.
\ No newline at end of file
diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 7696243..520234e 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -25,9 +25,9 @@ jobs:
           fi
           echo "release_version=$VERSION" >> "$GITHUB_OUTPUT"
 
-  # ── Job 1: Build .mcpb for Claude Desktop ───────────────────────────────────
+  # ── Job 1: Build .mcpb for Claude Desktop + standalone MCP zip ─────────────
   build-mcpb:
-    name: Build .mcpb
+    name: Build MCP Packages
     needs: validate-release-version
     runs-on: ubuntu-latest
     permissions:
@@ -76,15 +76,38 @@ jobs:
             mv "$f" "bindery-${VERSION}.mcpb" 2>/dev/null || true
           done
 
+      - name: Package standalone MCP server zip
+        run: |
+          VERSION="$RELEASE_VERSION"
+          mkdir -p mcpb/standalone-mcp/server
+          cp mcpb/server/index.js mcpb/standalone-mcp/server/index.js
+          cp mcpb/standalone-mcp-README.md mcpb/standalone-mcp/README.md
+          cd mcpb/standalone-mcp
+          zip "../../bindery-mcp-server-${VERSION}.zip" README.md server/index.js
+
+      - name: Render standardized release notes
+        run: |
+          VERSION="$RELEASE_VERSION"
+          sed "s/{{VERSION}}/${VERSION}/g" .github/release-notes-template.md > .github/release-notes.generated.md
+
       - name: Attest .mcpb build provenance
         uses: actions/attest-build-provenance@v4
         with:
           subject-path: mcpb/*.mcpb
 
+      - name: Attest standalone MCP zip provenance
+        uses: actions/attest-build-provenance@v4
+        with:
+          subject-path: bindery-mcp-server-*.zip
+
       - name: Upload .mcpb to release
         uses: softprops/action-gh-release@v3
         with:
-          files: mcpb/*.mcpb
+          files: |
+            mcpb/*.mcpb
+            bindery-mcp-server-*.zip
+          body_path: .github/release-notes.generated.md
+          generate_release_notes: true
           fail_on_unmatched_files: true
 
   # ── Job 2: Package VS Code extension + attach VSIX to release ───────────────
diff --git a/README.md b/README.md
index 043e413..3693219 100644
--- a/README.md
+++ b/README.md
@@ -85,7 +85,10 @@ Then in Obsidian: Settings → Community plugins (if not restricted) → Install
 
 ### [mcpb/](mcpb/) — Claude Desktop Extension
 
-Packages the MCP server as a `.mcpb` file for one-click installation in Claude Desktop or Cowork.
+Release assets now include two MCP package formats:
+
+- `bindery-<version>.mcpb` for Claude Desktop / Cowork
+- `bindery-mcp-server-<version>.zip` as a standalone MCP package for clients that support command + args + env setup (for example ChatGPT Work or LM Studio)
 
 **Download the latest release** from [Releases](../../releases) — no build step needed.
 
@@ -110,6 +113,21 @@ Packages the MCP server as a `.mcpb` file for one-click installation in Claude D
    - **Note:** full embedding can be a heavy operation, depending on your hardware, when running a local Ollama instance.
 6. Tools are now available — the agent calls `list_books` to discover book names
 
+### Standalone MCP Clients (ChatGPT Work, LM Studio, etc.)
+
+1. Download `bindery-mcp-server-*.zip` from the [latest release](../../releases/latest)
+2. Unzip it to a stable folder
+3. In your MCP client, add an MCP server with:
+  - eg. in ChatGPT Work -> **Settings** -> **Plug-ins** -> **MCP's** tab
+  - Command: `node`
+  - Args: absolute path to `server/index.js` from the unzipped folder
+4. Set environment variables:
+  - `BINDERY_BOOKS` (required): semicolon-separated `Name=path` pairs
+  - `BINDERY_OLLAMA_URL` (optional): Ollama endpoint for semantic features
+  - `BINDERY_ENABLE_SEMANTIC_INDEX` (optional): `true` to enable full semantic index builds
+  - `BINDERY_DEFAULT_SEARCH_MODE` (optional): `lexical`, `semantic_rerank`, or `full_semantic`
+5. Save and reconnect the MCP server
+
 ## Architecture Overview
 
 The following flow shows how Bindery's writing environments connect to agents, MCP, and skills at a high level.
diff --git a/mcpb/README.md b/mcpb/README.md
index cc00cd7..dc31f0c 100644
--- a/mcpb/README.md
+++ b/mcpb/README.md
@@ -1,4 +1,4 @@
-# Bindery MCP — Desktop Extension (.mcpb)
+# Bindery MCP — Desktop Packages
 
 Book authoring tools for Claude Desktop: chapter navigation, full-text search,
 translation management, session memory, typography formatting, and version snapshots.
@@ -20,7 +20,7 @@ Works with any Markdown book project structured with the Bindery VS Code extensi
 - **Version snapshots** — git-based save points after writing sessions
 - **Review diffs** — structured git diff of uncommitted changes
 
-## Manual Installation
+## Manual Installation (Claude Desktop)
 
 To install manually without using published Claude Connectors
 
@@ -29,6 +29,21 @@ To install manually without using published Claude Connectors
 3. Click **Install from file** (or drag-drop the `.mcpb` file)
 4. Fill in the **Books** field (see Configuration below)
 
+## Manual Installation (Standalone MCP clients)
+
+1. Download the `bindery-mcp-server-*.zip` file from the latest release
+2. Unzip it to a stable folder
+3. In your MCP client, add a new MCP server.
+   - eg. in ChatGPT Work -> **Settings** -> **Plug-ins** -> **MCP's** tab
+4. With:
+   - Command: `node`
+   - Args: absolute path to `server/index.js` from the unzipped folder
+5. Add environment variables:
+   - `BINDERY_BOOKS` (required): semicolon-separated `Name=path` pairs
+   - `BINDERY_OLLAMA_URL` (optional): Ollama endpoint for semantic features
+   - `BINDERY_ENABLE_SEMANTIC_INDEX` (optional): `true` to enable full semantic index builds
+   - `BINDERY_DEFAULT_SEARCH_MODE` (optional): `lexical`, `semantic_rerank`, or `full_semantic`
+
 ## Configuration
 
 | Setting | Required | Description |
@@ -165,7 +180,7 @@ table with issue type, location, and the reference that contradicts it.
 | `identify_book` | Match a working directory to a book name |
 | `health` | Server status: settings, index, embedding backend |
 | `init_workspace` | Create or update `.bindery/settings.json`, `translations.json`, generated `.bindery/README.md`, and the opinionated Arc / Notes / Characters / SESSION / PREFERENCES / memory / status scaffold |
-| `setup_ai_files` | Generate AI instruction files (CLAUDE.md, copilot-instructions.md, .cursor/rules, AGENTS.md), Claude skill templates, skill zips, and refresh generated `.bindery/README.md` |
+| `setup_ai_files` | Generate AI instruction files (CLAUDE.md, copilot-instructions.md, .cursor/rules, AGENTS.md), Claude skill templates, and refresh generated `.bindery/README.md` |
 | `index_build` | Build or rebuild the lexical index and, when enabled, the semantic embedding index |
 | `index_status` | Show lexical and semantic index status, build times, and stale hints |
 | `get_text` | Read any file by relative path, with optional line range |
@@ -230,6 +245,8 @@ mcpb pack
 
 This produces `bindery-mcp-<version>.mcpb`.
 
+For standalone MCP clients, release CI also produces `bindery-mcp-server-<version>.zip` containing `server/index.js` and setup instructions.
+
 ## Files
 
 - `manifest.json` — Extension metadata, tool declarations, user config schema
diff --git a/mcpb/standalone-mcp-README.md b/mcpb/standalone-mcp-README.md
new file mode 100644
index 0000000..211e597
--- /dev/null
+++ b/mcpb/standalone-mcp-README.md
@@ -0,0 +1,38 @@
+# Bindery Standalone MCP Server Package
+
+This package contains:
+
+- `server/index.js` (the bundled Bindery MCP server)
+- `README.md` (this file)
+
+## Install in any MCP client
+
+1. Unzip this archive to a stable folder on your machine.
+2. In your MCP client, add a new MCP server.
+   - eg. in ChatGPT Work -> **Settings** -> **Plug-ins** -> **MCP's** tab
+3. Use the following values:
+   - Command: `node`
+   - Args: absolute path to `server/index.js` in the folder you unzipped
+4. Configure environment variables (at minimum `BINDERY_BOOKS`).
+
+## Environment variables
+
+- `BINDERY_BOOKS` (required)
+  - Semicolon-separated `Name=path` pairs.
+  - Example: `MyBook=C:\Users\Me\MyBook;MyNovel=D:\Writing\MyNovel`
+- `BINDERY_OLLAMA_URL` (optional)
+  - Ollama endpoint for semantic reranking/full semantic search.
+  - Example: `http://localhost:11434`
+- `BINDERY_ENABLE_SEMANTIC_INDEX` (optional)
+  - `true` to enable full semantic index builds via `index_build`.
+  - Default: `false`
+- `BINDERY_DEFAULT_SEARCH_MODE` (optional)
+  - One of `lexical`, `semantic_rerank`, `full_semantic`.
+  - Default: `lexical`
+
+## Notes
+
+- Keep the unzipped folder in place after configuring the MCP server.
+- If Node is not on PATH, use the absolute path to the Node executable as Command.
+- If Node is not installed at all you can get it from *https://nodejs.org/*
+- This package is suitable for any MCP client that supports executable command + arguments + environment variables (for example ChatGPT Work or LM Studio).