diff --git a/server/services/model-discovery.js b/server/services/model-discovery.js
new file mode 100644
index 0000000..c7d6628
--- /dev/null
+++ b/server/services/model-discovery.js
@@ -0,0 +1,120 @@
+/**
+ * Model Discovery Service
+ *
+ * Fetches available models from MODELS_API_BASE_URL/v1/models (9router or
+ * any OpenAI-compatible endpoint). Falls back to hardcoded constants when
+ * the endpoint is unavailable or not configured.
+ *
+ * Cache TTL: 5 minutes. Re-fetched on first request after expiry.
+ */
+
+import { CLAUDE_MODELS } from '../../shared/modelConstants.js';
+
+const CACHE_TTL_MS = 5 * 60 * 1000; // 5 minutes
+
+let cachedModels = null;
+let cacheExpiry = 0;
+
+/**
+ * Maps a raw /v1/models entry to the { value, label } shape used by the UI.
+ * Filters to models that are relevant for Claude routing (id contains 'claude'
+ * or starts with 'cc/').
+ *
+ * @param {Object} entry - Raw model object from /v1/models
+ * @returns {{ value: string, label: string } | null}
+ */
+function mapModelEntry(entry) {
+  const id = typeof entry?.id === 'string' ? entry.id.trim() : null;
+  if (!id) return null;
+
+  // Only surface Claude-family and cc/* (9router) models for the Claude provider.
+  // Extend this filter if you want to surface all models.
+  const isClaude = id.toLowerCase().includes('claude') || id.startsWith('cc/');
+  if (!isClaude) return null;
+
+  // Build a human-readable label from the id.
+  // e.g. "cc/claude-sonnet-4-6" → "claude-sonnet-4-6 (9router)"
+  //      "claude-3-5-sonnet-20241022" → "claude-3-5-sonnet-20241022"
+  let label = entry.name ?? id;
+  if (id.startsWith('cc/')) {
+    label = `${id.slice(3)} (9router)`;
+  }
+
+  return { value: id, label };
+}
+
+/**
+ * Fetches models from the configured MODELS_API_BASE_URL/v1/models endpoint.
+ * Returns null on failure (caller uses fallback).
+ *
+ * @returns {Promise<Array<{value: string, label: string}> | null>}
+ */
+async function fetchModelsFromApi() {
+  const baseUrl = process.env.MODELS_API_BASE_URL || process.env.ANTHROPIC_BASE_URL;
+  if (!baseUrl) return null;
+
+  // Strip trailing /v1 if present so we can append /v1/models cleanly.
+  const cleanBase = baseUrl.replace(/\/v1\/?$/, '');
+  const url = `${cleanBase}/v1/models`;
+
+  try {
+    const apiKey = process.env.ANTHROPIC_API_KEY || '';
+    const res = await fetch(url, {
+      headers: {
+        Authorization: `Bearer ${apiKey}`,
+        'Content-Type': 'application/json',
+      },
+      signal: AbortSignal.timeout(8000),
+    });
+
+    if (!res.ok) {
+      console.warn(`[model-discovery] GET ${url} returned ${res.status}`);
+      return null;
+    }
+
+    const json = await res.json();
+    // OpenAI-compatible: { data: [...] }  or  { models: [...] }  or  plain array
+    const raw = Array.isArray(json) ? json : (json.data ?? json.models ?? []);
+    if (!Array.isArray(raw)) return null;
+
+    const mapped = raw.map(mapModelEntry).filter(Boolean);
+    if (mapped.length === 0) return null;
+
+    console.log(`[model-discovery] Loaded ${mapped.length} Claude models from ${url}`);
+    return mapped;
+  } catch (err) {
+    console.warn(`[model-discovery] Failed to fetch ${url}:`, err?.message ?? err);
+    return null;
+  }
+}
+
+/**
+ * Returns the list of available Claude model options.
+ * Uses cached results when fresh; falls back to CLAUDE_MODELS.OPTIONS.
+ *
+ * @returns {Promise<Array<{value: string, label: string}>>}
+ */
+export async function getAvailableClaudeModels() {
+  const now = Date.now();
+  if (cachedModels && now < cacheExpiry) {
+    return cachedModels;
+  }
+
+  const discovered = await fetchModelsFromApi();
+  if (discovered && discovered.length > 0) {
+    cachedModels = discovered;
+    cacheExpiry = now + CACHE_TTL_MS;
+    return cachedModels;
+  }
+
+  // Fallback: hardcoded list from shared constants
+  return CLAUDE_MODELS.OPTIONS;
+}
+
+/**
+ * Invalidates the model cache, forcing re-fetch on next call.
+ */
+export function invalidateModelCache() {
+  cachedModels = null;
+  cacheExpiry = 0;
+}