diff --git a/crates/sprout-cli/src/commands/messages.rs b/crates/sprout-cli/src/commands/messages.rs
index 53643070c..4d8f90ab2 100644
--- a/crates/sprout-cli/src/commands/messages.rs
+++ b/crates/sprout-cli/src/commands/messages.rs
@@ -5,8 +5,9 @@ use uuid::Uuid;
 use crate::client::SproutClient;
 use crate::error::CliError;
 use crate::validate::{
-    infer_language, normalize_mention_pubkeys, percent_encode, read_or_stdin, truncate_diff,
-    validate_content_size, validate_hex64, validate_uuid, MAX_DIFF_BYTES,
+    extract_at_names, infer_language, merge_mentions, normalize_mention_pubkeys, percent_encode,
+    read_or_stdin, truncate_diff, validate_content_size, validate_hex64, validate_uuid,
+    MAX_DIFF_BYTES,
 };
 
 // ---------------------------------------------------------------------------
@@ -60,6 +61,39 @@ async fn resolve_channel_id(client: &SproutClient, event_id: &str) -> Result<Uui
     )))
 }
 
+/// Resolve @names in content against channel members.
+/// Returns matching pubkeys. On any error, returns empty vec — never blocks a send.
+async fn resolve_content_mentions(
+    client: &SproutClient,
+    channel_id: &str,
+    content: &str,
+) -> Vec<String> {
+    let names = extract_at_names(content);
+    if names.is_empty() {
+        return vec![];
+    }
+    let body = client
+        .get_raw(&format!("/api/channels/{channel_id}/members"))
+        .await
+        .unwrap_or_default();
+    let parsed: serde_json::Value = serde_json::from_str(&body).unwrap_or_default();
+    let Some(members) = parsed["members"].as_array() else {
+        return vec![];
+    };
+    let mut pubkeys = Vec::new();
+    for m in members {
+        let Some(dn) = m["display_name"].as_str() else {
+            continue;
+        };
+        if names.iter().any(|n| n.eq_ignore_ascii_case(dn)) {
+            if let Some(pk) = m["pubkey"].as_str() {
+                pubkeys.push(pk.to_ascii_lowercase());
+            }
+        }
+    }
+    pubkeys
+}
+
 // ---------------------------------------------------------------------------
 // Read commands (unchanged)
 // ---------------------------------------------------------------------------
@@ -157,9 +191,11 @@ pub async fn cmd_send_message(
         None
     };
 
-    // Normalize mentions
-    let normalized: Vec<String> = normalize_mention_pubkeys(mentions, "");
-    let mention_refs: Vec<&str> = normalized.iter().map(|s| s.as_str()).collect();
+    // Normalize explicit mentions, then merge auto-resolved up to SDK cap of 50.
+    let mut merged: Vec<String> = normalize_mention_pubkeys(mentions, "");
+    let auto_resolved = resolve_content_mentions(client, channel_id, content).await;
+    merge_mentions(&mut merged, &auto_resolved, 50);
+    let mention_refs: Vec<&str> = merged.iter().map(|s| s.as_str()).collect();
 
     let builder = sprout_sdk::build_message(
         channel_uuid,
diff --git a/crates/sprout-cli/src/validate.rs b/crates/sprout-cli/src/validate.rs
index aa3920683..2c9c4dbcc 100644
--- a/crates/sprout-cli/src/validate.rs
+++ b/crates/sprout-cli/src/validate.rs
@@ -115,6 +115,65 @@ pub fn infer_language(file_path: &str) -> Option<String> {
     Some(lang.to_string())
 }
 
+/// Extract @mention names from message content.
+/// Returns lowercased names found after `@` tokens.
+/// Only matches `@word` preceded by whitespace or start-of-string.
+/// Characters allowed in names: alphanumeric, `.`, `-`, `_`.
+pub fn extract_at_names(content: &str) -> Vec<String> {
+    if content.is_empty() || !content.contains('@') {
+        return vec![];
+    }
+    let mut names: Vec<String> = Vec::new();
+    let mut seen = std::collections::HashSet::new();
+    let chars: Vec<char> = content.chars().collect();
+    let len = chars.len();
+    let mut i = 0;
+    while i < len {
+        if chars[i] == '@' {
+            // Must be at start-of-string or preceded by whitespace
+            let preceded_by_ws = i == 0 || chars[i - 1].is_ascii_whitespace();
+            if preceded_by_ws && i + 1 < len {
+                // Capture the name token: [a-zA-Z0-9._-]+
+                let start = i + 1;
+                let mut end = start;
+                while end < len {
+                    let c = chars[end];
+                    if c.is_ascii_alphanumeric() || c == '.' || c == '-' || c == '_' {
+                        end += 1;
+                    } else {
+                        break;
+                    }
+                }
+                if end > start {
+                    let name: String = chars[start..end].iter().collect();
+                    let lower = name.to_ascii_lowercase();
+                    if seen.insert(lower.clone()) {
+                        names.push(lower);
+                    }
+                }
+            }
+        }
+        i += 1;
+    }
+    names
+}
+
+/// Merge auto-resolved pubkeys into an explicit mention list, up to `cap`.
+/// Explicit mentions have priority; auto-resolved are added only if not already present.
+pub fn merge_mentions(explicit: &mut Vec<String>, auto_resolved: &[String], cap: usize) {
+    let budget = cap.saturating_sub(explicit.len());
+    let mut added = 0usize;
+    for pk in auto_resolved {
+        if added >= budget {
+            break;
+        }
+        if !explicit.contains(pk) {
+            explicit.push(pk.clone());
+            added += 1;
+        }
+    }
+}
+
 /// Normalize mention pubkeys: lowercase, deduplicate, remove sender's own pubkey.
 pub fn normalize_mention_pubkeys(pubkeys: &[String], sender_pubkey: &str) -> Vec<String> {
     let sender = sender_pubkey.to_ascii_lowercase();
@@ -335,6 +394,32 @@ mod tests {
         );
     }
 
+    // --- extract_at_names ---
+
+    #[test]
+    fn extract_at_names_matches() {
+        assert_eq!(extract_at_names("hello @alice"), vec!["alice"]);
+        assert_eq!(extract_at_names("@bob hello"), vec!["bob"]);
+        assert_eq!(
+            extract_at_names("@alice and @alice, meet @Bob"),
+            vec!["alice", "bob"]
+        );
+        assert_eq!(extract_at_names("line1\n@tyler line2"), vec!["tyler"]);
+        assert_eq!(
+            extract_at_names("@john.doe @mary_jane @bob-smith"),
+            vec!["john.doe", "mary_jane", "bob-smith"]
+        );
+    }
+
+    #[test]
+    fn extract_at_names_rejects() {
+        assert!(extract_at_names("").is_empty());
+        assert!(extract_at_names("no mentions").is_empty());
+        assert!(extract_at_names("user@example.com").is_empty());
+        assert!(extract_at_names("hello @ world").is_empty());
+        assert!(extract_at_names("hello @").is_empty());
+    }
+
     // --- normalize_mention_pubkeys ---
 
     #[test]
@@ -379,4 +464,30 @@ mod tests {
         let result = normalize_mention_pubkeys(&[], "sender");
         assert!(result.is_empty());
     }
+
+    // --- merge_mentions ---
+
+    #[test]
+    fn merge_mentions_dedup_and_cap() {
+        // basic merge
+        let mut m = vec!["a".into()];
+        merge_mentions(&mut m, &["b".into()], 50);
+        assert_eq!(m, ["a", "b"]);
+
+        // dedup: "a" already present
+        let mut m = vec!["a".into()];
+        merge_mentions(&mut m, &["a".into(), "b".into()], 50);
+        assert_eq!(m, ["a", "b"]);
+
+        // cap: 49 explicit + 2 auto → only 1 added
+        let mut m: Vec<String> = (0..49).map(|i| format!("{i:064}")).collect();
+        merge_mentions(&mut m, &["x".into(), "y".into()], 50);
+        assert_eq!(m.len(), 50);
+        assert_eq!(m.last().unwrap(), "x");
+
+        // at cap: 50 explicit → nothing added
+        let mut m: Vec<String> = (0..50).map(|i| format!("{i:064}")).collect();
+        merge_mentions(&mut m, &["extra".into()], 50);
+        assert_eq!(m.len(), 50);
+    }
 }
diff --git a/crates/sprout-mcp/src/server.rs b/crates/sprout-mcp/src/server.rs
index 1a23fa5f9..c3ff2ead7 100644
--- a/crates/sprout-mcp/src/server.rs
+++ b/crates/sprout-mcp/src/server.rs
@@ -71,6 +71,82 @@ fn find_root_from_tags(tags: &serde_json::Value) -> Option<String> {
 /// Maximum allowed content size for a single message (64 KiB).
 const MAX_CONTENT_BYTES: usize = 65_536;
 
+/// Extract @mention names from message content.
+/// Returns lowercased names found after `@` tokens.
+/// Only matches `@word` preceded by whitespace or start-of-string.
+/// Characters allowed in names: alphanumeric, `.`, `-`, `_`.
+fn extract_at_names(content: &str) -> Vec<String> {
+    if content.is_empty() || !content.contains('@') {
+        return vec![];
+    }
+    let mut names: Vec<String> = Vec::new();
+    let mut seen = std::collections::HashSet::new();
+    let chars: Vec<char> = content.chars().collect();
+    let len = chars.len();
+    let mut i = 0;
+    while i < len {
+        if chars[i] == '@' {
+            // Must be at start-of-string or preceded by whitespace
+            let preceded_by_ws = i == 0 || chars[i - 1].is_ascii_whitespace();
+            if preceded_by_ws && i + 1 < len {
+                // Capture the name token: [a-zA-Z0-9._-]+
+                let start = i + 1;
+                let mut end = start;
+                while end < len {
+                    let c = chars[end];
+                    if c.is_ascii_alphanumeric() || c == '.' || c == '-' || c == '_' {
+                        end += 1;
+                    } else {
+                        break;
+                    }
+                }
+                if end > start {
+                    let name: String = chars[start..end].iter().collect();
+                    let lower = name.to_ascii_lowercase();
+                    if seen.insert(lower.clone()) {
+                        names.push(lower);
+                    }
+                }
+            }
+        }
+        i += 1;
+    }
+    names
+}
+
+/// Resolve @names in content against channel members.
+/// Returns matching pubkeys. On any error, returns empty vec — never blocks a send.
+async fn resolve_content_mentions(
+    client: &RelayClient,
+    channel_id: &str,
+    content: &str,
+) -> Vec<String> {
+    let names = extract_at_names(content);
+    if names.is_empty() {
+        return vec![];
+    }
+    let body = client
+        .get(&format!("/api/channels/{channel_id}/members"))
+        .await
+        .unwrap_or_default();
+    let parsed: serde_json::Value = serde_json::from_str(&body).unwrap_or_default();
+    let Some(members) = parsed["members"].as_array() else {
+        return vec![];
+    };
+    let mut pubkeys = Vec::new();
+    for m in members {
+        let Some(dn) = m["display_name"].as_str() else {
+            continue;
+        };
+        if names.iter().any(|n| n.eq_ignore_ascii_case(dn)) {
+            if let Some(pk) = m["pubkey"].as_str() {
+                pubkeys.push(pk.to_ascii_lowercase());
+            }
+        }
+    }
+    pubkeys
+}
+
 /// Parameters for the `send_message` tool.
 #[derive(Debug, Serialize, Deserialize, schemars::JsonSchema)]
 pub struct SendMessageParams {
@@ -757,13 +833,32 @@ Default kind is 9 (stream message)."
         let kind_num = p
             .kind
             .unwrap_or(sprout_core::kind::KIND_STREAM_MESSAGE as u16);
-        let mention_refs: Vec<&str> = p
+        // Collect explicit pubkeys, dedup case-insensitively.
+        let mut seen = std::collections::HashSet::new();
+        let mut mentions: Vec<String> = p
             .mention_pubkeys
             .as_deref()
             .unwrap_or(&[])
             .iter()
-            .map(String::as_str)
+            .map(|s| s.to_ascii_lowercase())
+            .filter(|s| seen.insert(s.clone()))
             .collect();
+
+        // Auto-resolve @names in content and merge, up to SDK cap of 50.
+        let auto = resolve_content_mentions(&self.client, &p.channel_id, &p.content).await;
+        let budget = 50usize.saturating_sub(mentions.len());
+        let mut added = 0usize;
+        for pk in &auto {
+            if added >= budget {
+                break;
+            }
+            if !mentions.contains(pk) {
+                mentions.push(pk.clone());
+                added += 1;
+            }
+        }
+
+        let mention_refs: Vec<&str> = mentions.iter().map(String::as_str).collect();
         let broadcast = p.broadcast_to_channel.unwrap_or(false);
 
         // Build the event builder via SDK, routing by kind.
@@ -2431,6 +2526,34 @@ mod tests {
         assert_eq!(parsed.event_id, params.event_id);
         assert!(matches!(parsed.direction, VoteDirection::Up));
     }
+
+    // ── extract_at_names ──────────────────────────────────────────────────────
+
+    #[test]
+    fn extract_at_names_matches() {
+        // basic, start-of-string, dedup, newline, dots/hyphens/underscores
+        assert_eq!(extract_at_names("Hello @Tyler"), vec!["tyler"]);
+        assert_eq!(extract_at_names("@Tyler are you there?"), vec!["tyler"]);
+        assert_eq!(
+            extract_at_names("Hey @Alice and @alice, meet @Bob"),
+            vec!["alice", "bob"]
+        );
+        assert_eq!(extract_at_names("first line\n@Tyler second"), vec!["tyler"]);
+        assert_eq!(
+            extract_at_names("@john.doe @mary_jane @bob-smith"),
+            vec!["john.doe", "mary_jane", "bob-smith"]
+        );
+    }
+
+    #[test]
+    fn extract_at_names_rejects() {
+        // empty, no @, email, bare @, @ at EOF
+        assert!(extract_at_names("").is_empty());
+        assert!(extract_at_names("no mentions").is_empty());
+        assert!(extract_at_names("user@example.com").is_empty());
+        assert!(extract_at_names("hello @ world").is_empty());
+        assert!(extract_at_names("hello @").is_empty());
+    }
 }
 
 #[cfg(test)]