diff --git a/server/e2e/health_test.go b/server/e2e/health_test.go
index fbe1328..3366bf4 100644
--- a/server/e2e/health_test.go
+++ b/server/e2e/health_test.go
@@ -95,6 +95,17 @@ func TestHealthStandalone(t *testing.T) {
 	if got := grpcHealth(t, s); got != healthpb.HealthCheckResponse_SERVING {
 		t.Errorf("grpc health: %v", got)
 	}
+
+	// The metrics, with the connection just made.
+	code, metrics := monitorGet(t, s, "/_metrics")
+	if code != 200 {
+		t.Fatalf("/_metrics: %d %s", code, metrics)
+	}
+	for _, want := range []string{"unitdb_build_info{", "unitdb_up_seconds ", `unitdb_dependency_up{dep="store"} 1`, "unitdb_connections "} {
+		if !strings.Contains(metrics, want) {
+			t.Errorf("/_metrics has no %q:\n%s", want, metrics)
+		}
+	}
 }
 
 func TestHealthCluster(t *testing.T) {
diff --git a/server/internal/cluster_health.go b/server/internal/cluster_health.go
index bc2148f..134817d 100644
--- a/server/internal/cluster_health.go
+++ b/server/internal/cluster_health.go
@@ -2,6 +2,7 @@ package internal
 
 import (
 	"fmt"
+	"sort"
 	"sync/atomic"
 	"time"
 )
@@ -57,3 +58,43 @@ func (c *Cluster) readiness() (bool, string) {
 	return true, fmt.Sprintf("in the ring: %d of %d nodes; leader %s, %s ago",
 		len(ring), configured, leader, age.Round(time.Millisecond))
 }
+
+// writeMetrics writes this node's view of the cluster: only what is cheap
+// and safe to read per scrape. Peers are labelled by their node names, a
+// fixed set.
+func (c *Cluster) writeMetrics(m *metricsWriter) {
+	b := func(v bool) float64 {
+		if v {
+			return 1
+		}
+		return 0
+	}
+	m.one("unitdb_cluster_nodes", "gauge", "Nodes in the cluster's configuration, this one included.", float64(len(c.nodes)+1))
+	m.one("unitdb_cluster_members", "gauge", "Nodes in the ring this node routes by.", float64(len(c.getRingNodes())))
+	m.one("unitdb_cluster_ring_version", "gauge", "The ring version last seen from the leader; 0 before any.", float64(c.clusterRing.Load()))
+	m.one("unitdb_cluster_rebuilding", "gauge", "Whether this node is copying its topics from the others.", b(c.rebuilding.Load()))
+	m.one("unitdb_cluster_leaving", "gauge", "Whether this node is leaving the cluster.", b(c.leaving.Load()))
+	if last := c.health.lastLeader.Load(); last != 0 {
+		m.one("unitdb_cluster_leader_age_seconds", "gauge", "Seconds since this node last heard from a leader, or was one.", time.Since(time.Unix(0, last)).Seconds())
+	}
+	c.pendingMu.Lock()
+	pending := len(c.pending)
+	c.pendingMu.Unlock()
+	m.one("unitdb_cluster_pending_hints", "gauge", "Replicas waiting in memory for a peer to take them.", float64(pending))
+
+	names := make([]string, 0, len(c.nodes))
+	for name := range c.nodes {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	m.head("unitdb_replication_queue", "gauge", "Replicas queued for a peer.")
+	withoutTLS := 0
+	for _, name := range names {
+		n := c.nodes[name]
+		m.value("unitdb_replication_queue", "peer="+quote(name), float64(len(n.repl)))
+		if caps, ok := n.capabilities(); !ok || !containsNode(caps.Capabilities, capTLS) {
+			withoutTLS++
+		}
+	}
+	m.one("unitdb_cluster_peers_without_tls", "gauge", "Peers that don't advertise TLS for cluster traffic.", float64(withoutTLS))
+}
diff --git a/server/internal/health.go b/server/internal/health.go
index b3e5b67..b20ec88 100644
--- a/server/internal/health.go
+++ b/server/internal/health.go
@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
 	"net"
 	"net/http"
 	"sort"
@@ -27,7 +28,8 @@ import (
 //     reads the results the checks left, and never waits on them.
 //   - /_status: the same, as JSON, with each check's detail.
 //
-// The gRPC server also answers grpc.health.v1, SERVING while ready.
+// The gRPC server also answers grpc.health.v1, SERVING while ready, and
+// /_metrics has the server's metrics (metrics.go).
 //
 // Each check runs every healthInterval in the background, with a
 // healthTimeout, and keeps its last result.
@@ -57,6 +59,8 @@ type healthMonitor struct {
 	draining atomic.Bool
 	grpc     *health.Server
 	srv      *http.Server
+	// metrics writes /_metrics, when set.
+	metrics func(io.Writer)
 }
 
 func newHealthMonitor(started time.Time) *healthMonitor {
@@ -170,6 +174,14 @@ func (h *healthMonitor) handler() http.Handler {
 		}
 		w.Write([]byte("ready"))
 	})
+	mux.HandleFunc("/_metrics", func(w http.ResponseWriter, r *http.Request) {
+		if h.metrics == nil {
+			http.NotFound(w, r)
+			return
+		}
+		w.Header().Set("content-type", "text/plain; version=0.0.4")
+		h.metrics(w)
+	})
 	mux.HandleFunc("/_status", func(w http.ResponseWriter, r *http.Request) {
 		why := h.notReady()
 		status := "ready"
diff --git a/server/internal/metrics.go b/server/internal/metrics.go
new file mode 100644
index 0000000..7692f44
--- /dev/null
+++ b/server/internal/metrics.go
@@ -0,0 +1,136 @@
+package internal
+
+import (
+	"fmt"
+	"io"
+	"runtime/debug"
+	"sort"
+	"strings"
+	"time"
+)
+
+// Metrics, in the Prometheus text format, at /_metrics on the monitor port:
+// what /varz computes (connections, subscriptions, messages and bytes, and
+// the event durations), the health checks, and in a cluster this node's
+// place in it. Labels stay bounded: never a client or a topic.
+
+// buildInfo is the version and commit the binary was built from, when go
+// recorded them.
+var buildInfo = func() (version, commit string) {
+	version, commit = "unknown", "unknown"
+	bi, ok := debug.ReadBuildInfo()
+	if !ok {
+		return
+	}
+	if bi.Main.Version != "" {
+		version = bi.Main.Version
+	}
+	for _, s := range bi.Settings {
+		if s.Key == "vcs.revision" && s.Value != "" {
+			commit = s.Value
+			if len(commit) > 12 {
+				commit = commit[:12]
+			}
+		}
+	}
+	return
+}
+
+type metricsWriter struct {
+	w    io.Writer
+	seen map[string]bool
+}
+
+// head writes a metric's HELP and TYPE once.
+func (m *metricsWriter) head(name, typ, help string) {
+	if m.seen[name] {
+		return
+	}
+	m.seen[name] = true
+	fmt.Fprintf(m.w, "# HELP %s %s\n# TYPE %s %s\n", name, help, name, typ)
+}
+
+func (m *metricsWriter) value(name, labels string, v float64) {
+	if labels != "" {
+		labels = "{" + labels + "}"
+	}
+	fmt.Fprintf(m.w, "%s%s %s\n", name, labels, formatFloat(v))
+}
+
+func (m *metricsWriter) one(name, typ, help string, v float64) {
+	m.head(name, typ, help)
+	m.value(name, "", v)
+}
+
+func formatFloat(v float64) string {
+	s := fmt.Sprintf("%g", v)
+	if strings.Contains(s, "e+") {
+		s = fmt.Sprintf("%.0f", v)
+	}
+	return s
+}
+
+func quote(s string) string {
+	r := strings.NewReplacer(`\`, `\\`, `"`, `\"`, "\n", `\n`)
+	return `"` + r.Replace(s) + `"`
+}
+
+// writeMetrics writes the server's metrics.
+func (s *_Service) writeMetrics(w io.Writer) {
+	m := &metricsWriter{w: w, seen: map[string]bool{}}
+
+	version, commit := buildInfo()
+	m.head("unitdb_build_info", "gauge", "The build: its version and commit.")
+	m.value("unitdb_build_info", "version="+quote(version)+",commit="+quote(commit), 1)
+	m.one("unitdb_up_seconds", "gauge", "Seconds since the server started.", time.Since(s.start).Seconds())
+
+	if s.meter != nil {
+		m.one("unitdb_connections", "gauge", "Client connections open.", float64(s.meter.Connections.Count()))
+		m.one("unitdb_subscriptions", "gauge", "Subscriptions held.", float64(s.meter.Subscriptions.Count()))
+		m.one("unitdb_messages_in_total", "counter", "Messages published to this server.", float64(s.meter.InMsgs.Count()))
+		m.one("unitdb_messages_out_total", "counter", "Messages delivered to subscribers.", float64(s.meter.OutMsgs.Count()))
+		m.one("unitdb_bytes_in_total", "counter", "Payload bytes published to this server.", float64(s.meter.InBytes.Count()))
+		m.one("unitdb_bytes_out_total", "counter", "Payload bytes delivered to subscribers.", float64(s.meter.OutBytes.Count()))
+
+		// How long handling each client packet took (hdl_conn.go), in nanoseconds.
+		ts := s.meter.ConnTimeSeries.Snapshot()
+		const name = "unitdb_event_duration_seconds"
+		m.head(name, "summary", "Time to handle a client's packet, from the meter's sample.")
+		for _, q := range []struct {
+			q string
+			v int64
+		}{{"0.5", int64(ts.P50())}, {"0.75", int64(ts.P75())}, {"0.95", int64(ts.P95())}, {"0.99", int64(ts.P99())}, {"0.999", int64(ts.P999())}} {
+			m.value(name, "quantile="+quote(q.q), time.Duration(q.v).Seconds())
+		}
+	}
+
+	if s.health != nil {
+		m.head("unitdb_dependency_up", "gauge", "Whether a health check passes (1) or fails (0).")
+		m.head("unitdb_dependency_check_seconds", "gauge", "How long a health check took, last time.")
+		s.health.mu.RLock()
+		names := make([]string, 0, len(s.health.results))
+		for n := range s.health.results {
+			names = append(names, n)
+		}
+		sort.Strings(names)
+		for _, n := range names {
+			r := s.health.results[n]
+			up := 0.0
+			if r.OK {
+				up = 1
+			}
+			m.value("unitdb_dependency_up", "dep="+quote(n), up)
+			m.value("unitdb_dependency_check_seconds", "dep="+quote(n), float64(r.LatencyMs)/1000)
+		}
+		s.health.mu.RUnlock()
+		draining := 0.0
+		if s.health.draining.Load() {
+			draining = 1
+		}
+		m.one("unitdb_draining", "gauge", "Whether the server is shutting down.", draining)
+	}
+
+	if c := Globals.Cluster; c != nil {
+		c.writeMetrics(m)
+	}
+}
diff --git a/server/internal/metrics_test.go b/server/internal/metrics_test.go
new file mode 100644
index 0000000..b63f632
--- /dev/null
+++ b/server/internal/metrics_test.go
@@ -0,0 +1,71 @@
+package internal
+
+import (
+	"bytes"
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestMetrics(t *testing.T) {
+	s := &_Service{start: time.Now().Add(-time.Minute), meter: NewMeter()}
+	defer s.meter.UnregisterAll()
+	s.meter.Connections.Inc(3)
+	s.meter.InMsgs.Inc(7)
+	s.meter.ConnTimeSeries.AddTime(2 * time.Millisecond)
+	s.health = newHealthMonitor(s.start)
+	s.health.add("store", func() (string, error) { return "open", nil })
+	s.health.runAll()
+
+	var b bytes.Buffer
+	s.writeMetrics(&b)
+	out := b.String()
+	for _, want := range []string{
+		"# TYPE unitdb_build_info gauge",
+		"unitdb_connections 3",
+		"unitdb_messages_in_total 7",
+		`unitdb_event_duration_seconds{quantile="0.5"}`,
+		`unitdb_dependency_up{dep="store"} 1`,
+		"unitdb_draining 0",
+	} {
+		if !strings.Contains(out, want) {
+			t.Errorf("no %q in:\n%s", want, out)
+		}
+	}
+	// Each metric's HELP and TYPE once, and every sample line well formed.
+	for _, line := range strings.Split(strings.TrimSpace(out), "\n") {
+		if strings.HasPrefix(line, "#") {
+			continue
+		}
+		if f := strings.Fields(line); len(f) != 2 || !strings.HasPrefix(f[0], "unitdb_") {
+			t.Errorf("bad sample line %q", line)
+		}
+	}
+	if n := strings.Count(out, "# TYPE unitdb_dependency_up "); n != 1 {
+		t.Errorf("TYPE of unitdb_dependency_up %d times", n)
+	}
+}
+
+func TestClusterMetrics(t *testing.T) {
+	c := &Cluster{thisNodeName: "a", nodes: map[string]*ClusterNode{
+		"b": {name: "b", repl: make(chan replicaItem, 4)},
+		"c": {name: "c", repl: make(chan replicaItem, 4)},
+	}}
+	c.ringNodes = []string{"a", "b"}
+	c.health.leaderSeen("b")
+	var b bytes.Buffer
+	c.writeMetrics(&metricsWriter{w: &b, seen: map[string]bool{}})
+	out := b.String()
+	for _, want := range []string{
+		"unitdb_cluster_nodes 3",
+		"unitdb_cluster_members 2",
+		`unitdb_replication_queue{peer="b"} 0`,
+		`unitdb_replication_queue{peer="c"} 0`,
+		"unitdb_cluster_peers_without_tls 2",
+		"unitdb_cluster_leader_age_seconds",
+	} {
+		if !strings.Contains(out, want) {
+			t.Errorf("no %q in:\n%s", want, out)
+		}
+	}
+}
diff --git a/server/internal/service.go b/server/internal/service.go
index a9aff5e..c76c48d 100644
--- a/server/internal/service.go
+++ b/server/internal/service.go
@@ -213,6 +213,7 @@ func (s *_Service) listen(addr string) {
 	// The health checks, and grpc.health.v1 on the gRPC server.
 	s.health = newHealthMonitor(s.start)
 	s.health.addServiceChecks()
+	s.health.metrics = s.writeMetrics
 	s.grpc.Register = func(g *grpc.Server) {
 		healthpb.RegisterHealthServer(g, s.health.grpc)
 	}