Revert optimization tool from task-based to synchronous

No MCP clients currently support the experimental Tasks API:
- MCP Inspector: task support not available
- Claude Desktop/CLI: returns error -32601 (task augmentation required)
- Goose: falls back to other tools

Changes:
- Remove task-based OptimizationTask.ts implementation
- Re-enable synchronous OptimizationTool.ts (with polling)
- Remove task capabilities from server (taskStore, InMemoryTaskStore, etc.)
- Update OptimizeDeliveriesPrompt to remove task-based language
- Tool now blocks for 5-10 seconds during optimization (necessary trade-off)

The synchronous version:
- Works in all current MCP clients
- Still does async polling internally (POST job, GET results)
- Returns results when optimization completes
- All 422 tests passing

We can convert back to task-based when clients add support.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

Author: mattpodwysocki

src/index.ts

@@ -15,17 +15,11 @@ import {
   ListPromptsRequestSchema,
   GetPromptRequestSchema
 } from '@modelcontextprotocol/sdk/types.js';
-import {
-  InMemoryTaskStore,
-  InMemoryTaskMessageQueue
-} from '@modelcontextprotocol/sdk/experimental/tasks/stores/in-memory.js';
 import { parseToolConfigFromArgs, filterTools } from './config/toolConfig.js';
 import { getAllTools } from './tools/toolRegistry.js';
 import { getAllResources } from './resources/resourceRegistry.js';
 import { getAllPrompts, getPromptByName } from './prompts/promptRegistry.js';
 import { getVersionInfo } from './utils/versionUtils.js';
-import { registerOptimizationTask } from './tools/optimization-tool/OptimizationTask.js';
-import { httpRequest } from './utils/httpPipeline.js';
 import {
   initializeTracing,
   shutdownTracing,
@@ -68,9 +62,6 @@ const enabledTools = filterTools(allTools, config);
 // Get all resources
 const allResources = getAllResources();
 
-// Create task store for async operations
-const taskStore = new InMemoryTaskStore();
-
 // Create an MCP server
 const server = new McpServer(
   {
@@ -81,26 +72,16 @@ const server = new McpServer(
     capabilities: {
       tools: {},
       resources: {},
-      prompts: {},
-      tasks: { requests: { tools: { call: {} } } }
-    },
-    taskStore,
-    taskMessageQueue: new InMemoryTaskMessageQueue()
+      prompts: {}
+    }
   }
 );
 
 // Register enabled tools to the server
 enabledTools.forEach((tool) => {
-  // Skip OptimizationTool as it's registered as a task instead
-  if (tool.name === 'optimization_tool') {
-    return;
-  }
   tool.installTo(server);
 });
 
-// Register task-based tools
-registerOptimizationTask(server, httpRequest);
-
 // Register all resources to the server
 allResources.forEach((resource) => {
   resource.installTo(server);

src/prompts/OptimizeDeliveriesPrompt.ts

@@ -63,17 +63,14 @@ Please follow these steps:
 2. Use optimization_tool to find the optimal route:
    - Pass coordinates array
    - Set profile to mapbox/${mode}
-   - The tool will run as an async task - you'll get a task ID immediately
-3. Poll the task until it completes (the tool handles this automatically)
-4. Once results are available, display:
+   - The tool will process the optimization (this may take 5-10 seconds for complex routes)
+3. Once results are available, display:
    - Optimized sequence of stops (which location to visit in what order)
    - Total distance and estimated travel time
    - Estimated arrival time at each stop
    - Map visualization showing the optimized route
    - Any locations that couldn't be included (dropped items)
 
-IMPORTANT: The optimization_tool is task-based and runs asynchronously. The tool will submit the job and poll for results in the background. Present the results once the task completes.
-
 Format the output to be clear with:
 - Numbered list of stops in optimal order
 - ETAs at each location