@nooticr/mcp 1.26.28 → 1.26.30

package/dist/shared/tools.js

@@ -14,7 +14,8 @@ import { OUTPUT_SCHEMAS, anyObject } from "./output-schemas.js";
 import { deriveHashtags, derivedNote } from "./hashtags.js";
 import { createTaskStore, registerSlowTool } from "./tasks.js";
 import { COMMENT_CATEGORIES, COMMENT_SENTIMENTS, platformFromUrl, reviewGuidance, toEvidence, } from "./comment-review.js";
-import { costSentence, EVIDENCE_PLANS, fetchBillingNote, frameIndex, framesToBlocks, planCalls, scoreDraftGuidance, } from "./evidence.js";
+import { costSentence, EVIDENCE_PLANS, fetchBillingNote, frameIndex, framesToBlocks, planCalls, scoreDraftGuidance, KNOWN_PLATFORMS, platformFailureGuidance, } from "./evidence.js";
+import { evidenceDigest } from "./evidence-digest.js";
 import { confirmSpend, declinedResult, searchMentionsCost, SEARCH_PLATFORMS, DISCOVERABLE_PLATFORMS, } from "./spend.js";
 import { BackendWatchStore, MemoryWatchStore, registerWatchlist, } from "./watchlist.js";
 import { registerJobTools } from "./jobs.js";
@@ -25,7 +26,7 @@ import { loadingPlansJson } from "./loading-plans.js";
 import { registerHandoff } from "./handoff.js";
 import { registerCollabTools } from "./collab.js";
 /** Current MCP server version — bumped on every deploy for traceability. */
-export const MCP_SERVER_VERSION = "1.26.28";
+export const MCP_SERVER_VERSION = "1.26.30";
 /** MCP Apps extension identifier */
 const UI_EXTENSION = "io.modelcontextprotocol/ui";
 /** MIME type for MCP Apps HTML resources */
@@ -342,6 +343,21 @@ function toolError(prefix, err) {
             isError: true,
         };
     }
+    // A network that ERRORED is not a network that came back empty, and the two
+    // want opposite responses: an empty result means try a different query, an
+    // error means try a different network or tell the user. Nothing said which,
+    // so a host did the one thing that cannot work — re-ran the same call with
+    // the words changed, three times, each a paid upstream call. The backend
+    // names the platform in these messages ("... failed: reddit: ..."), so the
+    // split is recoverable here without every one of the 26 call sites having to
+    // pass its arguments down.
+    const platform = KNOWN_PLATFORMS.find((p) => new RegExp(`(^|[^a-z])${p}([^a-z]|$)`, "i").test(msg));
+    if (platform) {
+        return {
+            content: [{ type: "text", text: `${prefix}. ${platformFailureGuidance({ platform, message: msg })}` }],
+            isError: true,
+        };
+    }
     return { content: [{ type: "text", text: `${prefix}: ${msg}` }], isError: true };
 }
 /**
@@ -377,16 +393,34 @@ async function runEvidence(tool, args, client) {
             extra = { unavailable: plan.also.via };
         }
     }
-    // The guidance leads. A model reads the first text block in a result, and
-    // that is where the account of what to produce has to be — everything below
-    // it is material with no instruction attached.
+    // The guidance rides in BOTH channels, and the structured one is the one
+    // that arrives.
+    //
+    // This used to say "the guidance leads: a model reads the first text block in
+    // a result, and that is where the account of what to produce has to be". That
+    // is what the MCP spec allows, and it is not what hosts do. Claude Code
+    // discards every `content` text block whenever a result carries
+    // `structuredContent`, keeps the non-text blocks, and shows the model the
+    // serialised JSON instead — unconditionally, with no setting, and without
+    // consulting `outputSchema`. Every tool here declares one, so every guidance
+    // string this file writes was landing nowhere.
+    //
+    // Measured before changing it, over 59 real runs of the quest suite: 0
+    // guidance phrases across 175 tool results. `analyze_post` handed back frames,
+    // a transcript and 2,000 characters ending in "call show_analysis when you are
+    // done"; the model wrote the analysis and delivered it in chat, and
+    // show_analysis was called in 1 run of 12.
+    //
+    // The text block stays for hosts that honour it (and for the images, which
+    // must sit in `content`). `guidance` on the payload is what reaches the rest.
+    // See docs/testing/tool-chaining-quests.md.
     const billing = nothingToFetch
         ? "Nothing was fetched for this call, so nothing was charged."
         : fetchBillingNote(tool);
-    const content = [
-        { type: "text", text: `${plan.guidance(args)}\n\n${billing}` },
-    ];
+    const guidance = `${plan.guidance(args)}\n\n${billing}`;
+    const content = [{ type: "text", text: guidance }];
     const out = {
+        guidance,
         // Kept, and constant. It is not an echo of an argument — the job tools in
         // jobs.ts set the same key with no argument to echo — it marks a payload as
         // material the caller still has to read. Dropping it would break anything
@@ -411,6 +445,19 @@ async function runEvidence(tool, args, client) {
     // rather than "no fetch was made".
     if (plan.also && !nothingToFetch)
         out[plan.also.via] = extra;
+    // The evidence has to ride in the text block too, and it is written last
+    // because `out` is what it renders — the fetched payload, plus whatever the
+    // `also` call added, minus the frames that became image blocks.
+    //
+    // Same reason the guidance is in both channels (#44/#51), pointed the other
+    // way: Claude Code drops these text blocks and reads `structuredContent`,
+    // while a host rendering a UI view hands the model the text blocks and gives
+    // `structuredContent` to the widget. A transcript that exists only in the
+    // second one is a transcript the model was told to read and never saw (#59).
+    const digest = evidenceDigest(out);
+    if (digest) {
+        content[0] = { type: "text", text: `${guidance}\n\n---\n\n${digest}` };
+    }
     return { content, structuredContent: out };
 }
 export function createMcpServer(rawMakeClient, opts) {
@@ -525,8 +572,8 @@ export function createMcpServer(rawMakeClient, opts) {
         // a search_mentions result render in the monitoring view.
         "answer_my_audience",
         "show_audience_replies",
-        "track_competitor",
-        // The comparison track_competitor computes per creator and never puts side
+        "track_creator",
+        // The comparison track_creator computes per creator and never puts side
         // by side, plus the free view that draws the table once a read is written.
         "compare_creators",
         "watchlist_standings",
@@ -538,6 +585,10 @@ export function createMcpServer(rawMakeClient, opts) {
         // hits rather than comments, so it gets its own view rather than the
         // monitoring one.
         "search_spoken_mentions",
+        // Prospect discovery: the same keyword search as discover_social_posts, but
+        // widened into the shapes a complaint takes rather than the topic it is
+        // about. Returns `posts`, so the gallery view draws it.
+        "find_people_with_problem",
         // Own-account intelligence (own-account.ts). list_own_apps stays
         // view-less like watch_creator — it lists metadata, nothing to draw.
         // get_scheduled_posts and get_post_performance return a `posts` array
@@ -907,7 +958,11 @@ export function createMcpServer(rawMakeClient, opts) {
             "Work out their niche, recurring themes, hook formula, what over- and under-performs and who " +
             "their audience is, reading the spread of the numbers rather than only the best post, and " +
             `name the posts you reason from. ${costSentence("analyze_creator_profile")} ` +
-            "Use for the teardown itself; find_hook_pattern fetches the same posts and asks only for the formula.",
+            "Use for the teardown itself; find_hook_pattern fetches the same posts and asks only for the " +
+            "formula. NOT for keeping an eye on a rival over time: \"track X\", \"how are they doing " +
+            "lately\", \"what have they shipped since I last looked\" are track_creator, which scores " +
+            "each post against that creator's own median and remembers where you left off. This tool " +
+            "returns the posts and no verdict, and remembers nothing between calls.",
         _meta: {
             ui: { resourceUri: uiResource("analyze_creator_profile") },
             "ui/resourceUri": uiResource("analyze_creator_profile"),
@@ -1160,19 +1215,32 @@ export function createMcpServer(rawMakeClient, opts) {
             const res = await client.callTool("get_post_comments", { ...args });
             const structured = (res.structured ?? {});
             const comments = toEvidence(args.url, structured.comments);
+            const guidance = reviewGuidance(args.url, comments.length);
+            // "Here are 4 comments ... Read them and classify each one" was a false
+            // sentence in the channel it was written in: the comments were only ever
+            // in `structuredContent`, which a host rendering the view keeps for the
+            // widget. The model was asked to classify comments it could not see, and
+            // to return the `id` of each — so the digest renders ids too (#59).
+            const payload = {
+                // See runEvidence: a text block alone reaches no host that renders
+                // structuredContent, and this is the tool show_comment_review's whole
+                // chain hangs off.
+                guidance,
+                mode: "evidence",
+                url: args.url,
+                platform: structured.platform ?? null,
+                commentCount: comments.length,
+                comments,
+                // Passed through rather than re-derived: the platform's own clustering
+                // is evidence too, and it is already paid for.
+                themes: structured.themes ?? [],
+                mcpCredits: structured.mcpCredits ?? null,
+            };
             return {
-                content: [{ type: "text", text: reviewGuidance(args.url, comments.length) }],
-                structuredContent: {
-                    mode: "evidence",
-                    url: args.url,
-                    platform: structured.platform ?? null,
-                    commentCount: comments.length,
-                    comments,
-                    // Passed through rather than re-derived: the platform's own clustering
-                    // is evidence too, and it is already paid for.
-                    themes: structured.themes ?? [],
-                    mcpCredits: structured.mcpCredits ?? null,
-                },
+                content: [
+                    { type: "text", text: `${guidance}\n\n---\n\n${evidenceDigest(payload)}` },
+                ],
+                structuredContent: payload,
             };
         }
         catch (err) {
@@ -1921,16 +1989,24 @@ export function createMcpServer(rawMakeClient, opts) {
     // argument: the draft is the caller's own text, so the backend was only ever
     // being paid to hold it up against a standard, and the standard is what
     // comes back instead. Nothing here can fail, which is why nothing is caught.
-    async (args) => ({
-        content: [{ type: "text", text: scoreDraftGuidance(args.draft, String(args.platform ?? "")) }],
-        structuredContent: {
-            draft: args.draft,
-            platform: args.platform ?? "tiktok",
-            // Same shape the free tools use, so a caller totting up a session's spend
-            // does not have to special-case this one.
-            mcpCredits: { cost: 0 },
-        },
-    }));
+    async (args) => {
+        const guidance = scoreDraftGuidance(args.draft, String(args.platform ?? ""));
+        return {
+            content: [{ type: "text", text: guidance }],
+            structuredContent: {
+                // The rubric IS this tool's answer — it fetches nothing and scores
+                // nothing itself. In a text block it reached no host that renders
+                // structuredContent, which made the tool return its own input and a
+                // credit note. See runEvidence.
+                guidance,
+                draft: args.draft,
+                platform: args.platform ?? "tiktok",
+                // Same shape the free tools use, so a caller totting up a session's spend
+                // does not have to special-case this one.
+                mcpCredits: { cost: 0 },
+            },
+        };
+    });
     server.registerTool("repurpose_post", {
         title: "Repurpose Post",
         description: "The source post, its transcript and its stats, for you to rewrite for other surfaces — X " +
@@ -2247,7 +2323,7 @@ export function createMcpServer(rawMakeClient, opts) {
         }
     });
     registerPrompts(server);
-    // One store for both. track_competitor keeps its "since I last looked" marker
+    // One store for both. track_creator keeps its "since I last looked" marker
     // on the same watchlist entries, in its own field — two stores would mean a
     // creator you watch and a creator you track were different people.
     //

package/dist/shared/ui-template.js

@@ -2782,7 +2782,7 @@ export const NOOTICR_UI_TEMPLATE = `<!DOCTYPE html>
   /**
    * The one line a competitor sweep or an underperformance check actually
    * exists to produce — ratio to the creator's own median, and the verdict
-   * that ratio earns. track_competitor and why_did_this_underperform attach
+   * that ratio earns. track_creator and why_did_this_underperform attach
    * "standing" to every post they score (see performance.ts), and until this
    * existed postCard rendered the same view/like/comment pills any gallery
    * gets, so the computed comparison a person paid for never reached the
@@ -3982,8 +3982,18 @@ export const NOOTICR_UI_TEMPLATE = `<!DOCTYPE html>
 
     // Single post
     if(d.post||d.platform){app.innerHTML=postCard(d.post||d,true);return;}
-    // Fallback
-    app.innerHTML='<div class="json-block fade-in">'+esc(JSON.stringify(d,null,2))+"</div>";
+    // Fallback. The guidance field is prose written for the model, not data —
+    // dumped into the JSON block it reads as a wall of escaped text, and it is
+    // the longest string in most payloads. Drawn above the block and dropped
+    // from it. This is also what finally gives score_draft a view worth
+    // looking at: its rubric IS its answer, and the card used to show the
+    // echoed draft and a credit cost with the rubric nowhere on screen.
+    var rest=d,lead="";
+    if(d&&typeof d==="object"&&typeof d.guidance==="string"){
+      lead='<div class="fade-in" style="white-space:pre-wrap;line-height:1.5;margin-bottom:12px">'+esc(d.guidance)+"</div>";
+      rest={};for(var gk in d){if(gk!=="guidance")rest[gk]=d[gk];}
+    }
+    app.innerHTML=lead+'<div class="json-block fade-in">'+esc(JSON.stringify(rest,null,2))+"</div>";
     setTimeout(reportSize,50);
   }
 

package/dist/shared/watchlist.js

@@ -33,11 +33,18 @@
  * registers, and only one registration under a name can win — but renaming
  * either side would break whichever hosts had already learned the old one,
  * for a tool whose whole promise is "what you called before still works". So
- * nothing about the surface moved. `watch_creator`, `unwatch_creator`,
- * `catch_up_watchlist` and `track_competitor` keep their names, arguments and
+ * nothing about the surface moved for THAT change. `watch_creator`,
+ * `unwatch_creator` and `catch_up_watchlist` keep their names, arguments and
  * behaviour; only the bytes moved, and the backend's own twins stay
  * unregistered here because they would be a second, disconnected list.
  *
+ * `track_creator` is the one exception, and for an unrelated reason: it was
+ * `track_competitor`, and that name was measured to be unreachable for the
+ * way most people ask for it — see its registration in jobs.ts. The backend
+ * still registers `track_competitor` for its own dashboard copilot, which is
+ * the ordinary state of two repos with two release cycles rather than a
+ * collision: the names no longer overlap, so nothing has to win.
+ *
  * The stores above are still live and still matter. They are the fallback for
  * an account with no workspace or a backend too old to have the tools, and
  * they are what the one-time migration reads from. Deleting one because "the
@@ -155,9 +162,10 @@ export class FileWatchStore {
  * and on ChatGPT had two, and a server-side scheduler could read neither.
  *
  * The swap is deliberately at `WatchStore` rather than at the tool layer.
- * `watch_creator`, `unwatch_creator`, `catch_up_watchlist` and
- * `track_competitor` keep their names, their arguments and their behaviour;
- * only where the bytes live changes. That is the whole reason the collision
+ * `watch_creator`, `unwatch_creator` and `catch_up_watchlist` keep their
+ * names, their arguments and their behaviour; only where the bytes live
+ * changes. (`track_creator` was later renamed from `track_competitor` for a
+ * reason that has nothing to do with storage — see above.) That is the whole reason the collision
  * with the backend's identically-named tools was not worth solving by
  * renaming anything: a host that already learned `watch_creator` keeps
  * working, and it now writes somewhere the other host can see.

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@nooticr/mcp",
   "mcpName": "com.nooticr/mcp",
-  "version": "1.26.28",
+  "version": "1.26.30",
   "description": "MCP server for nooticr - fetch, discover and understand TikTok/Instagram/YouTube/X/Reddit/LinkedIn/Weibo posts with AI (media metadata, niche discovery, brand monitoring, hook & viral analysis)",
   "homepage": "https://mcp.nooticr.com",
   "repository": {
@@ -38,9 +38,10 @@
     "test:e2e:visual": "playwright test tests/e2e/agentic-visual.e2e.ts tests/e2e/agentic-visual-full-app.e2e.ts",
     "test:e2e-smoke": "bash scripts/run-mechanical-e2e-smoke.sh",
     "test:e2e-smoke:fixture": "NOOTICR_E2E_BACKEND=fixture bash scripts/run-mechanical-e2e-smoke.sh",
-    "test:agentic-e2e": "bash scripts/run-agentic-evals.sh",
+    "test:quests": "bash scripts/run-quests.sh",
     "contract:host": "sh scripts/host-contract.sh",
-    "verify": "npm run build && npx tsc --noEmit && (cd cloudflare && npx tsc --noEmit) && npx vitest run && npm run contract:host && npm run conformance:mcpjam && npx playwright test",
+    "chain:map": "bash scripts/run-chain-map.sh",
+    "verify": "npm run build && npx tsc --noEmit && (cd cloudflare && npx tsc --noEmit) && npx vitest run && npm run contract:host && npm run chain:map && npm run conformance:mcpjam && npx playwright test",
     "contract:manifest": "node scripts/contract-check.mjs --from=../nooticr-server/platform-capabilities.json",
     "test:mutation": "node scripts/mutation-guard.mjs",
     "test:invariants": "node scripts/invariant-guard.mjs",