feat(infra): token-saver MCP — both sides deployed

Local (qfox-1):
  - token_saver_mcp.py: 8 tools, free compute first, paid fallback
  - build_server.py: HTTP build server with /health on port 8765
  - lean-build-server.service: systemd user service (auto-start)
  - opencode.json: token-saver registered

Cloud bootstrap:
  - cloud_bootstrap.sh: installs elan, registers MCP, verifies connectivity

Architecture:
  Any MCP client -> token-saver MCP -> neon/qfox-1 (free, 0 tokens)
                                    -> OpenRouter (paid fallback, proofs only)

Build: 3571 jobs, 0 errors

5-Applications/tools-scripts/mcp/build_server.py

@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+"""Lean Build Server — listens on port 8765, runs `lake build <module>`.
+
+Endpoints:
+  GET /<module>   — build module, return JSON {passed, output}
+  GET /health     — return {"status": "ok", "modules": [...]}
+"""
+import subprocess, json, sys, os
+from http.server import HTTPServer, BaseHTTPRequestHandler
+
+LAKE_WORKDIR = os.environ.get("LAKE_WORKDIR",
+    "/home/allaun/Research Stack/0-Core-Formalism/lean/Semantics")
+
+def list_modules():
+    """Parse lakefile.toml for [[lean_lib]] names."""
+    import re
+    try:
+        text = open(f"{LAKE_WORKDIR}/lakefile.toml").read()
+        return re.findall(r'^name\s*=\s*"(\w+)"', text, re.MULTILINE)
+    except: return []
+
+class BuildHandler(BaseHTTPRequestHandler):
+    def do_GET(self):
+        path = self.path.lstrip("/") or "Semantics"
+        if path == "health":
+            resp = json.dumps({"status": "ok", "workdir": LAKE_WORKDIR,
+                               "modules": list_modules()}).encode()
+        else:
+            try:
+                result = subprocess.run(
+                    ["lake", "build", path],
+                    capture_output=True, text=True, timeout=600, cwd=LAKE_WORKDIR,
+                )
+                ok = result.returncode == 0
+                output = result.stdout + result.stderr
+            except subprocess.TimeoutExpired:
+                ok = False; output = "TIMEOUT"
+            resp = json.dumps({"passed": ok, "output": output[-2000:]}).encode()
+        self.send_response(200)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Access-Control-Allow-Origin", "*")
+        self.end_headers()
+        self.wfile.write(resp)
+
+    def log_message(self, format, *args):
+        sys.stderr.write("[build-server] %s\n" % (format % args))
+
+HTTPServer(("0.0.0.0", 8765), BuildHandler).serve_forever()

5-Applications/tools-scripts/mcp/cloud_bootstrap.sh

@@ -0,0 +1,64 @@
+#!/bin/bash
+# TokenSaver Cloud Bootstrap — run once per cloud session.
+# Registers the token-saver MCP so all token-burning operations route
+# to free local compute (neon CPU / qfox-1 build server).
+#
+# Environment variables (set these in cloud env config):
+#   BUILD_SERVER_URL=http://100.88.57.96:8765
+#   NEON_OLLAMA_URL=http://100.92.88.64:11434
+#   LAKE_WORKDIR=/workspace/0-Core-Formalism/lean/Semantics
+
+set -e
+
+echo "[token-saver] Installing Lean toolchain..."
+curl -sSf https://raw.githubusercontent.com/leanprover/elan/master/elan-init.sh | sh -s -- -y
+export PATH="$HOME/.elan/bin:$PATH"
+
+echo "[token-saver] Registering MCP server..."
+mkdir -p ~/.config/opencode
+
+# Merge token-saver into existing opencode config
+cat > /tmp/token_saver_mcp.json << 'EOF'
+{
+  "token-saver": {
+    "type": "local",
+    "command": ["python3", "5-Applications/tools-scripts/mcp/token_saver_mcp.py"],
+    "environment": {
+      "BUILD_SERVER_URL": "{env:BUILD_SERVER_URL}",
+      "NEON_OLLAMA_URL": "{env:NEON_OLLAMA_URL}",
+      "LAKE_WORKDIR": "{env:LAKE_WORKDIR}"
+    },
+    "enabled": true
+  }
+}
+EOF
+
+if [ -f ~/.config/opencode/opencode.jsonc ]; then
+  python3 -c "
+import json
+with open('/root/.config/opencode/opencode.jsonc') as f:
+    cfg = json.load(f)
+cfg.setdefault('mcp', {}).update(json.load(open('/tmp/token_saver_mcp.json')))
+with open('/root/.config/opencode/opencode.jsonc', 'w') as f:
+    json.dump(cfg, f, indent=2)
+"
+else
+  cp /tmp/token_saver_mcp.json ~/.config/opencode/opencode.jsonc
+fi
+
+echo "[token-saver] Verifying reachable resources..."
+python3 -c "
+import urllib.request
+try:
+    r = urllib.request.urlopen('${BUILD_SERVER_URL:-http://100.88.57.96:8765}/health', timeout=5)
+    print('  build_server: UP')
+except: print('  build_server: DOWN (tailnet required)')
+try:
+    r = urllib.request.urlopen('${NEON_OLLAMA_URL:-http://100.92.88.64:11434}/api/tags', timeout=5)
+    print('  neon_ollama:   UP')
+except: print('  neon_ollama:   DOWN (tailnet required)')
+"
+
+echo "[token-saver] Ready. MCP tools available:"
+echo "  verify_build, compile_check, linter_check, classify_proof,"
+echo "  generate_proof (neon → paid fallback), format_code, doc_lookup"