diff --git a/backend/__tests__/unit/config/nativeAgents.scoutKnowledge.test.js b/backend/__tests__/unit/config/nativeAgents.scoutKnowledge.test.js
index ab38f08955..dc744441ac 100644
--- a/backend/__tests__/unit/config/nativeAgents.scoutKnowledge.test.js
+++ b/backend/__tests__/unit/config/nativeAgents.scoutKnowledge.test.js
@@ -141,6 +141,29 @@ describe('Scout onboarding knowledge', () => {
     expect(prompt).toMatch(/and the MCP line/);
   });
 
+  // Stranger journey 2026-10-26: Scout wrote the Terminal steps and asked
+  // "Want me to open that page?" instead of proposing, and no card came.
+  test('proposes the connect card once it knows the tool and OS, without asking first', () => {
+    expect(prompt).toMatch(
+      /Propose the connect card in the same turn you learn both what they run and an OS where it can be connected/,
+    );
+    expect(prompt).toMatch(/Do not ask whether to: the card is the question/);
+    expect(prompt).toMatch(/Do not write the Terminal steps in chat instead of the card/);
+    expect(prompt).toMatch(
+      /Where the connect card applies, the card is the Terminal half: propose it and give only the MCP line/,
+    );
+  });
+
+  // A card gets only the MCP line in text, so Codex needs a line of its own:
+  // the Claude Code one does not configure Codex.
+  test('gives an MCP line for each tool the card connects', () => {
+    expect(prompt).toContain('`claude mcp add commonly -e COMMONLY_API_URL=... '
+      + '-e COMMONLY_AGENT_TOKEN=... -- npx -y @commonlyai/mcp`');
+    expect(prompt).toContain('`codex mcp add commonly --env COMMONLY_API_URL=... '
+      + '--env COMMONLY_AGENT_TOKEN=... -- npx -y @commonlyai/mcp`');
+    expect(prompt).toMatch(/Give the line for the tool they run/);
+  });
+
   test('cites developer docs that exist in the repository', () => {
     const repo = path.resolve(__dirname, '../../../..');
     for (const doc of [
diff --git a/backend/config/native-agents/scout.ts b/backend/config/native-agents/scout.ts
index 0c18feeae1..2f22c35e8b 100644
--- a/backend/config/native-agents/scout.ts
+++ b/backend/config/native-agents/scout.ts
@@ -104,10 +104,13 @@ export const scoutApp = {
     + 'tool gets a clear error instead of quietly starting the other one. That '
     + 'last command is what makes their agent answer @mentions, and its Terminal '
     + 'window must stay open. The page turns green when the agent first checks in.\n'
-    + '  * An MCP line for Claude Code (`claude mcp add commonly -e '
+    + '  * An MCP line gives their agent Commonly tools inside their own '
+    + 'sessions. For Claude Code: `claude mcp add commonly -e '
     + 'COMMONLY_API_URL=... -e COMMONLY_AGENT_TOKEN=... -- npx -y '
-    + '@commonlyai/mcp`) gives their agent Commonly tools inside their own '
-    + 'sessions. MCP alone does NOT answer @mentions; only the listener does. '
+    + '@commonlyai/mcp`. For Codex: `codex mcp add commonly --env '
+    + 'COMMONLY_API_URL=... --env COMMONLY_AGENT_TOKEN=... -- npx -y '
+    + '@commonlyai/mcp`. Give the line for the tool they run. '
+    + 'MCP alone does NOT answer @mentions; only the listener does. '
     + 'Say exactly this when someone set up MCP and asks why their agent is '
     + 'silent.\n'
     // Walked 2026-10-19 (CLI 0.1.108, OpenCode 1.19.3, macOS, a non-admin
@@ -203,6 +206,14 @@ export const scoutApp = {
     + 'to the page with the command and what it needs, so do not post the steps '
     + 'again; at most one short line that the command is on the page that opened. '
     + 'You never see their key and must never ask them to paste one into chat.\n'
+    // Stranger journey 2026-10-26 02:18Z: told "macOS" after "Claude Code",
+    // Scout pasted the listener steps and asked "Want me to open that page?",
+    // then waited. No card ever came, and a stranger who does not answer that
+    // question never gets one. The run 21 minutes earlier proposed on that turn.
+    + '- Propose the connect card in the same turn you learn both what they run '
+    + 'and an OS where it can be connected. Do not ask whether to: the card '
+    + 'is the question, and they answer it by approving. Do not write the '
+    + 'Terminal steps in chat instead of the card.\n'
     + '- commonly_agent_status tells you every agent\'s live state. Use it '
     + 'whenever the user asks whether an agent is connected or why it is '
     + 'silent: never-connected means the connect-page step was never finished '
@@ -226,7 +237,9 @@ export const scoutApp = {
     + '- When their own tool can call agents or command-line tools, or they '
     + 'ask about both directions, give both halves: the Terminal command (their agent '
     + 'answers @mentions here) and the MCP line (their agent uses Commonly '
-    + 'from its own sessions). Say that MCP alone does not answer @mentions.\n'
+    + 'from its own sessions). Say that MCP alone does not answer @mentions. '
+    + 'Where the connect card applies, the card is the Terminal half: propose '
+    + 'it and give only the MCP line in your text.\n'
     + '- End with one next step they can take right now.\n'
     + '- The facts above are verified: give them without hedging. For '
     + 'anything they do not cover, say you are not sure, give the closest '