@roadmapperai/mcp 0.5.0 → 0.7.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@roadmapperai/mcp",
-  "version": "0.5.0",
+  "version": "0.7.0",
   "description": "Roadmapper AI MCP server — exposes a planning surface (themes, capabilities, tasks, sprints, PRs) to coding agents via stdio JSON-RPC. Pairs with the Roadmapper AI workspace at dashboard.roadmapperai.com.",
   "keywords": [
     "mcp",

package/server.mjs

@@ -1090,7 +1090,8 @@ const TOOLS = [
       "USE WHEN: the user asks to plan features, design new work, sketch a roadmap, file a TODO that should persist beyond this conversation, or break a capability into deliverables.\n" +
       "PREREQUISITE: get_agents_md once this session (the server enforces this and returns an error with a `fix` field if missing). Call suggest_capability_for first to find the right parent capability — do not invent a new one.\n" +
       "ANTI-PATTERN: do not call to track in-progress work within a single conversation — use the harness TodoWrite tool. Do not call to log a bug discovered during implementation — file in the issue tracker, not roadmapper. Do not call when you don't know which capability the task belongs under; resolve that first.\n" +
-      "EXAMPLE: propose_task({ capabilityId: 'CAP-XXX', title: 'Drag-and-drop block reorder', acceptance: ['Block can be dragged with mouse + keyboard', 'Order persists across reloads'], idempotencyKey: 'session-1-task-3' })\n\n" +
+      "REQUIRED FIELDS: capabilityId, title, effort. Always size the task — XS (≤2h) / S (≤1d) / M (~1-3d) / L (~1-2w) / XL (>2w). Effort drives capability % roll-up weighting; do not omit.\n" +
+      "EXAMPLE: propose_task({ capabilityId: 'CAP-XXX', title: 'Drag-and-drop block reorder', effort: 'M', acceptance: ['Block can be dragged with mouse + keyboard', 'Order persists across reloads'], idempotencyKey: 'session-1-task-3' })\n\n" +
       "Requires SUPABASE_SERVICE_ROLE_KEY. Pass idempotencyKey so retries don't duplicate. Pass dryRun: true to validate without writing. Pass workspaceId to target a workspace other than the env default.",
     inputSchema: {
       type: "object",
@@ -1118,7 +1119,7 @@ const TOOLS = [
         dryRun: { type: "boolean" },
         workspaceId: { type: "string" },
       },
-      required: ["capabilityId", "title"],
+      required: ["capabilityId", "title", "effort"],
       additionalProperties: false,
     },
   },
@@ -1942,7 +1943,16 @@ async function proposeTask(args, projected, wsId) {
   if (!cap) return errorResult(`Capability ${args.capabilityId} not found.`);
   const titleErr = validateName(args.title, 5);
   if (titleErr) return errorResult(titleErr);
-  if (args.effort && !VALID_EFFORTS.has(args.effort))
+  // Effort is required end-to-end: the schema flags it, and the
+  // Postgres RPC refuses without it. We also raise here so older
+  // clients that ignore the schema get a clear error message before
+  // a round-trip.
+  if (!args.effort) {
+    return errorResult(
+      "effort is required (one of XS, S, M, L, XL). Always size the task — XS (≤2h), S (≤1d), M (~1-3d), L (~1-2w), XL (>2w)."
+    );
+  }
+  if (!VALID_EFFORTS.has(args.effort))
     return errorResult(`Invalid effort ${args.effort}.`);
   if (args.priority && !VALID_PRIORITIES.has(args.priority))
     return errorResult(`Invalid priority ${args.priority}.`);
@@ -1959,7 +1969,7 @@ async function proposeTask(args, projected, wsId) {
   )
     return errorResult(`expectedScope must be a positive number, got ${args.expectedScope}.`);
 
-  const effort = args.effort ?? "M";
+  const effort = args.effort;
   const start = todayISO();
   // Target dates are day-resolution; round up so sub-day estimates
   // (XS=0.25, S=0.5) still nudge the target at least one day out.
@@ -2268,10 +2278,32 @@ function suggestCapabilityFor(args, projected) {
     (c) => effectiveCapabilityStatus(c, projected.tasks) !== "delivered"
   );
   const query = tokenize(desc);
+  // Build a richer haystack per capability: include the parent
+  // theme's name+description (vocabulary often overlaps with the
+  // query) and the titles of attached tasks (the *implementation*
+  // language, which is closer to how engineers describe new work
+  // than the bet-style capability outcome statement). This makes
+  // matches that previously scored near zero hit the 0.2+ band
+  // where the model takes them seriously.
+  const themeById = new Map(
+    (projected.themes ?? []).map((t) => [t.id, t])
+  );
   const ranked = activeCaps
     .map((c) => {
+      const theme = themeById.get(c.pillarId);
+      const taskTitles = (projected.tasks ?? [])
+        .filter((t) => t.capabilityId === c.id)
+        .map((t) => t.title)
+        .join(" ");
       const hay = tokenize(
-        `${c.name} ${c.description ?? ""} ${c.outcome ?? ""}`
+        [
+          c.name,
+          c.description ?? "",
+          c.outcome ?? "",
+          theme?.name ?? "",
+          theme?.description ?? "",
+          taskTitles,
+        ].join(" ")
       );
       return { capability: c, score: jaccardScore(query, hay) };
     })