docs: reconcile docs to the tag-only release model; read version from package.json

.cursorrules

@@ -10,6 +10,6 @@ This is a mcp-server repository for Local AI MCP development.
 ## Conventions
 
 - Use conventional commits (feat:, fix:, chore:, docs:)
-- Never manually edit the version -- CI auto-bumps package.json
+- Bump the version in `package.json` in your PR; CI tags and publishes it
 - Keep mcp-tools.json in sync with the registered tools
 - Provider adapters live in src/providers/; tools live in src/tools/

AGENTS.md

@@ -19,10 +19,12 @@ This is an MCP server. It contains:
 ## Branching and commit model
 
 - **Single branch**: `main` only. No develop/release branches.
-- **Conventional commits** are required. The release workflow parses them:
-  - `feat:` or `feat(scope):` -- triggers a **minor** version bump
-  - `feat!:` or `BREAKING CHANGE` -- triggers a **major** version bump
-  - Everything else (`fix:`, `chore:`, `docs:`, etc.) -- triggers a **patch** bump
+- **Conventional commits** are required. Use them to decide your version bump:
+  - `feat:` or `feat(scope):` -- bump the **minor** version
+  - `feat!:` or a `BREAKING CHANGE` trailer -- bump the **major** version
+  - everything else (`fix:`, `chore:`, `docs:`, etc.) -- bump the **patch** version
+
+  Apply the bump in your PR with `npm version <patch|minor|major> --no-git-tag-version` (keeps `package.json` and the lockfile in sync); `release.yml` tags and publishes that version on merge. CI never writes to `main`.
 - Commit messages should be concise and describe the "why", not the "what".
 
 ## CI/CD workflows
@@ -66,7 +68,7 @@ Keeps repository labels in sync.
 ## Code conventions
 
 - No hardcoded credentials -- CI scans for password/token/api_key patterns.
-- Conventional commits; never hand-edit the version.
+- Conventional commits; bump the version deliberately in your PR (CI tags and publishes it).
 - Keep `mcp-tools.json` in sync with the tools registered in `src/tools/`.
 
 ## Adding content

CHANGELOG.md

@@ -1,23 +1,32 @@
 # Changelog
 
 All notable changes to Local AI MCP are documented in this file. The format is
-based on [Keep a Changelog](https://keepachangelog.com/). Versions are bumped
-automatically from conventional commits by the release workflow; never edit the
-version by hand.
+based on [Keep a Changelog](https://keepachangelog.com/). Bump the version in
+each PR (following conventional-commit intent); on merge the release workflow
+tags that version and publishes it.
 
 ## [Unreleased]
 
+## [0.1.2] - 2026-06-14
+
+### Changed
+
+- Release chain is now PAT-free: `release.yml` uses the default `GITHUB_TOKEN`
+  to push tags and create the GitHub Release, then explicitly dispatches
+  `publish.yml` (a `GITHUB_TOKEN` release does not auto-trigger it). Removed the
+  dependency on `RELEASE_PAT`, which was read-only and could not push.
+
 ## [0.1.1] - 2026-06-14
 
 ### Changed
 
 - Docs and metadata now describe an MCP server (provider adapters, tool
   registration, stdio transport, npx install) instead of cursor-plugin
   boilerplate.
-- Rewrote the release workflow to auto-bump from conventional commits and create
-  the release with a token that actually triggers publishing; the publish
-  workflow now runs on a published release or manual dispatch and skips a
-  version that is already on npm.
+- Adopted a tag-only release model compatible with the protected `main` branch:
+  bump the version in your PR; `release.yml` tags it and creates the release,
+  and `publish.yml` builds, tests, and publishes (skipping a version already on
+  npm). CI never writes to `main`.
 
 ### Fixed
 

CLAUDE.md

@@ -8,7 +8,7 @@ This file provides guidance for Claude Code when working in this repository.
 
 Local AI MCP -- Unified MCP server for managing local model runtimes (Ollama, LM Studio, and more): provider-agnostic discovery, lifecycle, hardware-fit, and delegated inference.
 
-**Version:** 0.1.0
+**Version:** 0.1.2
 **License:** CC-BY-NC-ND-4.0
 **Author:** TMHSDigital
 

CONTRIBUTING.md

@@ -32,7 +32,7 @@ Use [Conventional Commits](https://www.conventionalcommits.org/):
 - Add it to `mcp-tools.json`
 - Add vitest tests
 
-Never hand-edit the version; CI auto-bumps `package.json`.
+Bump the version in `package.json` in your PR (e.g. `npm version <patch|minor|major> --no-git-tag-version`); CI tags and publishes it on merge.
 
 ## Pull Request Process
 

src/index.ts

@@ -1,4 +1,7 @@
 #!/usr/bin/env node
+import { readFileSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
 import { loadConfig } from "./config.js";
@@ -7,13 +10,26 @@ import { ProviderManager } from "./providers/manager.js";
 import type { ToolContext } from "./tools/context.js";
 import { registerAll } from "./tools/index.js";
 
+// Read the version from package.json so the server reports the published
+// version rather than a hardcoded literal that drifts on every release.
+// dist/index.js lives at <pkg>/dist/, so ../package.json is the manifest
+// (npm always includes package.json in the published tarball).
+function readVersion(): string {
+  try {
+    const pkgPath = join(dirname(fileURLToPath(import.meta.url)), "..", "package.json");
+    return JSON.parse(readFileSync(pkgPath, "utf-8")).version ?? "0.0.0";
+  } catch {
+    return "0.0.0";
+  }
+}
+
 async function main(): Promise<void> {
   const config = loadConfig(process.env);
   const manager = new ProviderManager(config);
   const hardware = createHardwareProbe();
   const ctx: ToolContext = { manager, hardware, config };
 
-  const server = new McpServer({ name: "local-ai-mcp", version: "0.1.0" });
+  const server = new McpServer({ name: "local-ai-mcp", version: readVersion() });
   registerAll(server, ctx);
 
   const transport = new StdioServerTransport();