lockage 0.9.2 → 0.9.3

package/README.md

@@ -54,11 +54,11 @@ No clone or build is required. You only add a definition to your MCP client's co
 
 The first run may take a little time to fetch. The DB (ledger) is created at `~/.lockage/lockage.sqlite` by default, so no extra configuration is needed.
 
-To get started you need just two things: **registering with an MCP client** and **distributing the coordination rules** (next section). Line these two up and multi-agent coordination starts working.
+To get started you need just two things: **registering with an MCP client** and **distributing the coordination rules** (next section). Line these two up and multi-agent coordination starts working. For Claude Code and Antigravity, installing the plugin (next section) covers both.
 
 ### Registering with an MCP client
 
-Add a server definition like the following to the MCP configuration of Kiro / Claude Code / Antigravity, and so on (follow each client's configuration file format. Antigravity uses `~/.gemini/config/mcp_config.json`, and you can also register with `agy mcp add`).
+Add a server definition like the following to the MCP configuration of Kiro, and so on; Claude Code and Antigravity need no registration here (see below). Follow each client's configuration file format.
 
 To Kiro's `mcp.json`, add the following definition that launches the npm-published artifact with `npx` (no clone or `npm run build` is needed; the first run may take a little time to fetch).
 
@@ -74,14 +74,25 @@ To Kiro's `mcp.json`, add the following definition that launches the npm-publish
 }
 ```
 
-The DB (ledger) is stored at `~/.lockage/lockage.sqlite` by default (the parent directory is created automatically). The default is fine; only those who want to change the location add `"env": { "LOCKAGE_DB": "/absolute/path/to/your-workspace/.lockage/lockage.sqlite" }` to `lockage`.
+The DB (ledger) is stored at `~/.lockage/lockage.sqlite` by default (the parent directory is created automatically). The default is fine; only those who want to change the location add `"env": { "LOCKAGE_DB": "/absolute/path/to/your-workspace/.lockage/lockage.sqlite" }` to `lockage`. With the Claude Code plugin, set `LOCKAGE_DB` in the `env` of a Claude Code settings file (usually `~/.claude/settings.json`) instead (see the Claude Code section of [`plugins/README.md`](plugins/README.md#claude-code)). In Antigravity, set `LOCKAGE_DB` in the shell environment from which Antigravity is launched.
 
 When multiple agents coordinate, line everyone up so they **point at the same `LOCKAGE_DB`**. If you leave the default, everyone points at the same default ledger, so no extra configuration is needed. This becomes the shared coordination ledger.
 
-For clients such as Claude Code / Antigravity where launching from the npm-published artifact has not been confirmed, clone and `npm run build`, then register with the traditional definition that points `command:node`, `args:["/absolute/path/to/lockage/dist/index.js"]`, and `env.LOCKAGE_DB` (for Antigravity, `agy mcp add --env LOCKAGE_DB=<absolute path to the ledger> lockage node <path to the lockage repository>/dist/index.js`).
+For Claude Code and Antigravity, you do not need to register it here: the lockage plugin (next section) bundles this `npx -y lockage` definition, so installing the plugin registers the server too. If you registered it by hand before, remove that registration (with a different command, two lockage servers line up; with the same command, only the hand registration is kept and the bundled definition goes unused), after moving any `LOCKAGE_DB` in its `env` to the settings file (for Claude Code, usually `~/.claude/settings.json`) or environment variable (for Antigravity; see [`plugins/README.md`](plugins/README.md)).
 
 Do not write `LOCKAGE_TOOLSETS` in an agent's configuration (only the default `core` is visible). Only when a human uses the maintenance/diagnostic tools, add `"LOCKAGE_TOOLSETS": "maintenance"` to the `env` of that MCP client.
 
+With the Claude Code plugin, the bundled server has no `env`, so register a separate server for maintenance only where a human uses it, and remove it when done:
+
+```bash
+claude mcp add lockage-maint -s local -e LOCKAGE_TOOLSETS=maintenance -- npx -y lockage@latest
+claude mcp remove lockage-maint -s local
+```
+
+Give it a command different from the bundled one (here `lockage@latest`, which can be newer than the version the bundled server runs). Claude Code merges servers with the same command even when their names and `env` differ, so registering it with `npx -y lockage` would replace the bundled server, leaving only the maintenance server in that session (confirmed with Claude Code 2.1.285).
+
+This server exposes the core tools as well as the maintenance ones, as `mcp__lockage-maint__*` (31 tools with lockage 0.9.2, confirmed). While it is registered, lockage-worker sees a second lockage server and returns "failed", so do not launch workers until you remove it. It reads `LOCKAGE_DB` from the environment Claude Code passes to MCP servers, like the bundled server, so it points at the ledger set in the settings `env` (not tried for this server).
+
 ---
 
 ## Distributing the coordination rules
@@ -90,8 +101,8 @@ Once both registering with an MCP client and distributing these coordination rul
 
 **The main path is installation via a plugin / Power.** You can load the coordination rules and hooks on the client side without placing anything in the target repository.
 
-- **Claude Code**: register this repository as a marketplace and install the lockage plugin (`plugins/claude`).
-- **Antigravity**: install with `agy plugin install <local path where you cloned lockage>/plugins/antigravity`.
+- **Claude Code**: register this repository as a marketplace and install the lockage plugin (`plugins/claude`). The plugin also registers the MCP server.
+- **Antigravity**: install with `agy plugin install <local path where you cloned lockage>/plugins/antigravity`. The plugin also registers the MCP server.
 - **Kiro**: register `plugins/kiro` as a Power and place the hooks into the target's `.kiro/hooks/` (`npx -y lockage setup kiro`).
 
 For the detailed procedure of each, see [`plugins/README.md`](plugins/README.md).
@@ -102,7 +113,7 @@ For the detailed procedure of each, see [`plugins/README.md`](plugins/README.md)
 
 Merely registering this server in the MCP configuration does not tell an agent "when, which tool, in what order to call." Only once the instructions required for coordination (the coordination rules in `AGENTS.md` and the lockage-workflow skill) are delivered to each agent does coordination start running as the rules describe.
 
-> 💡 There are only two prerequisites for coordinating multiple agents. **(a)** Everyone's MCP configuration points at **the same absolute `LOCKAGE_DB` path**, and **(b)** each agent names **its own unique `agent_id`** (since a shared `AGENTS.md` cannot write them out in advance, each agent decides its own at startup and confirms uniqueness on the board). Line these two up and coordination runs as the rules describe.
+> 💡 There are only two prerequisites for coordinating multiple agents. **(a)** Everyone's lockage server points at **the same absolute `LOCKAGE_DB` path** (via the MCP configuration's `env`, or the environment the server inherits), and **(b)** each agent names **its own unique `agent_id`** (since a shared `AGENTS.md` cannot write them out in advance, each agent decides its own at startup and confirms uniqueness on the board). Line these two up and coordination runs as the rules describe.
 
 ### Mapping of the coordination flow to tools
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "lockage",
-  "version": "0.9.2",
+  "version": "0.9.3",
   "description": "Lockage — an MCP server for multiple coding LLMs to collaborate around a local git repository (single host)",
   "license": "MIT",
   "type": "module",