diff --git a/packages/cli-v3/src/dev/mcpServer.ts b/packages/cli-v3/src/dev/mcpServer.ts
index c63926e1d..8c4e57da3 100644
--- a/packages/cli-v3/src/dev/mcpServer.ts
+++ b/packages/cli-v3/src/dev/mcpServer.ts
@@ -1,11 +1,13 @@
+import polka from "polka";
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { SSEServerTransport } from "@modelcontextprotocol/sdk/server/sse.js";
-import { optional, union, z } from "zod";
+import { z } from "zod";
 import { logger } from "../utilities/logger.js";
 import { CliApiClient } from "../apiClient.js";
 import { ApiClient, RunStatus } from "@trigger.dev/core/v3";
-import polka from "polka";
+import { eventBus } from "../utilities/eventBus.js";
 
+let allTaskIds: string[] = [];
 let projectRef: string;
 let dashboardUrl: string;
 // there is some overlap between `ApiClient` and `CliApiClient` which is not ideal
@@ -20,6 +22,22 @@ const server = new McpServer({
   version: "1.0.0",
 });
 
+// The `list-all-tasks` tool primarily helps to enable fuzzy matching of task IDs (names).
+// This way, one doesn't need to specify the full task ID and rather let the LLM figure it out.
+// This could be a good fit for the `resource` entity in MCP.
+// Also, a custom `prompt` entity could be useful to instruct the LLM to prompt the user
+// for selecting a task from a list of matching tasks, when the confidence for an exact match is low.
+server.tool("list-all-tasks", "List all available task IDs in the worker.", async () => {
+  return {
+    content: [
+      {
+        text: JSON.stringify(allTaskIds, null, 2),
+        type: "text",
+      },
+    ],
+  };
+});
+
 server.tool(
   "trigger-task",
   "Trigger a task",
@@ -213,6 +231,10 @@ app.post("/messages", (req, res) => {
   }
 });
 
+eventBus.on("backgroundWorkerInitialized", (worker) => {
+  allTaskIds = worker.manifest?.tasks.map((task) => task.id) ?? [];
+});
+
 export const startMcpServer = async (options: {
   port: number;
   cliApiClient: CliApiClient;