abtestresult-mcp 1.0.8 → 1.1.0

package/README.md

@@ -2,7 +2,7 @@
 
 Statistical tools for A/B testing, available as a [Model Context Protocol](https://modelcontextprotocol.io) (MCP) server. Powered by [abtestresult.com](https://abtestresult.com).
 
-Give any AI assistant (Claude, Cursor, VS Code, etc.) the ability to analyze A/B tests, calculate sample sizes, run Bayesian analysis, and more. All calculations run locally on your machine — no API keys, your test data never leaves your device.
+Give any AI assistant (Claude, Codex, Cursor, VS Code, etc.) the ability to analyze A/B tests, calculate sample sizes, run Bayesian analysis, and more. All calculations run locally on your machine — no API keys, your test data never leaves your device.
 
 This package ships readable source. It is source-available for local personal use only and is not licensed for commercial use.
 
@@ -23,7 +23,8 @@ Add to your MCP configuration:
 
 **Where to add this:**
 
-- **Claude Desktop:** `~/.claude/claude_desktop_config.json`
+- **Claude Desktop:** `~/Library/Application Support/Claude/claude_desktop_config.json`
+- **Codex:** `~/.codex/config.toml`
 - **Claude Code:** `.mcp.json` in your project root
 - **Cursor:** `.cursor/mcp.json`
 - **VS Code:** `.vscode/settings.json` (use `mcp.servers` instead of `mcpServers`)
@@ -43,6 +44,14 @@ That's it. No API keys, no authentication. The server starts automatically when
 | `paired_test` | Paired T-test or Wilcoxon signed-rank for before/after data |
 | `survey_sample_size` | Cochran's formula with finite population correction |
 
+## Shareable result links
+
+Every analysis tool (except `paired_test`) returns a `view_url` in its response — a
+deeplink to [abtestresult.com](https://abtestresult.com) that pre-fills the matching
+calculator with your inputs and auto-runs it. Open it to inspect the result visually
+(charts, confidence intervals, recommendations) or share it with your team. No data is
+stored server-side: the inputs are encoded directly into the URL.
+
 ## Usage Examples
 
 Once connected, just ask your AI assistant in natural language:

package/dist/index.js

@@ -34719,9 +34719,29 @@ function trackToolUsage(toolName) {
 }
 var server = new McpServer({
   name: "ABTestResult",
-  version: "1.0.0",
+  version: "1.1.0",
   description: "Statistical tools for A/B testing \u2014 significance testing, sample size calculation, Bayesian analysis, and more. Powered by abtestresult.com"
 });
+var SITE_URL = (process.env.ABTESTRESULT_BASE_URL || "https://abtestresult.com").replace(/\/+$/, "");
+function toBase64Url(str) {
+  return Buffer.from(str, "utf-8").toString("base64url");
+}
+function buildDeeplink(path, compact) {
+  const clean = {};
+  for (const [k, v] of Object.entries(compact)) {
+    if (v !== void 0 && v !== null && v !== "") clean[k] = v;
+  }
+  return `${SITE_URL}${path}?d=${toBase64Url(JSON.stringify(clean))}`;
+}
+function pipeJoin(parts) {
+  const arr = parts.map((p) => p === void 0 || p === null ? "" : String(p));
+  while (arr.length > 0 && arr[arr.length - 1] === "") arr.pop();
+  return arr.join("|");
+}
+function tailCode(testType, niMargin) {
+  if (niMargin > 0) return "n";
+  return testType === "one-sided" ? "1" : "2";
+}
 server.tool(
   "analyze_ab_test",
   `Analyze an A/B test with rate/proportion metrics (e.g. conversion rate, click-through rate).
@@ -34752,11 +34772,22 @@ Example: "Control had 5000 visitors with 250 conversions, variant had 5000 visit
       params.non_inferiority_margin
     );
     const summary = result.isSignificant ? `\u2705 SIGNIFICANT \u2014 The variant ${result.delta > 0 ? "outperforms" : "underperforms"} the control.` : `\u23F3 NOT SIGNIFICANT \u2014 No statistically significant difference detected.`;
+    const ctrlRate = (params.control_conversions / params.control_users * 100).toFixed(2);
+    const varRate = (params.variant_conversions / params.variant_users * 100).toFixed(2);
+    const viewUrl = buildDeeplink("/statistical-significance-calculator", {
+      c: pipeJoin([params.control_users, params.control_conversions, ctrlRate]),
+      v: pipeJoin([params.variant_users, params.variant_conversions, varRate]),
+      m: "r",
+      t: tailCode(params.test_type, params.non_inferiority_margin),
+      l: String(Math.round(params.confidence_level * 100)),
+      n: params.non_inferiority_margin > 0 ? String(params.non_inferiority_margin) : ""
+    });
     return {
       content: [{
         type: "text",
         text: JSON.stringify({
           summary,
+          view_url: viewUrl,
           significant: result.isSignificant,
           p_value: round(result.pValue, 6),
           z_score: round(result.zScore, 4),
@@ -34817,11 +34848,20 @@ Example: "Control had 1000 users with mean $45.20 (std $12.50), variant had 1000
       params.non_inferiority_margin
     );
     const summary = result.isSignificant ? `\u2705 SIGNIFICANT \u2014 The variant ${result.delta > 0 ? "outperforms" : "underperforms"} the control.` : `\u23F3 NOT SIGNIFICANT \u2014 No statistically significant difference detected.`;
+    const viewUrl = buildDeeplink("/statistical-significance-calculator", {
+      c: pipeJoin([params.control_users, params.control_mean, "", params.control_std_dev]),
+      v: pipeJoin([params.variant_users, params.variant_mean, "", params.variant_std_dev]),
+      m: "a",
+      t: tailCode(params.test_type, params.non_inferiority_margin),
+      l: String(Math.round(params.confidence_level * 100)),
+      n: params.non_inferiority_margin > 0 ? String(params.non_inferiority_margin) : ""
+    });
     return {
       content: [{
         type: "text",
         text: JSON.stringify({
           summary,
+          view_url: viewUrl,
           significant: result.isSignificant,
           p_value: round(result.pValue, 6),
           t_score: round(result.tScore, 4),
@@ -34910,10 +34950,21 @@ Example: "How many users do I need to detect a 10% relative lift on a 5% baselin
         params.daily_traffic ?? null
       );
     }
+    const viewUrl = buildDeeplink("/sample-size-calculator", {
+      m: params.metric_type === "rate" ? "r" : "a",
+      t: params.test_type === "one-sided" ? "1" : "2",
+      b: String(params.baseline),
+      d: String(params.mde),
+      s: params.metric_type === "average" && params.std_dev ? String(params.std_dev) : "",
+      c: String(Math.round(params.confidence_level * 100)),
+      p: String(Math.round(params.power * 100)),
+      v: params.num_variants !== 1 ? String(params.num_variants) : ""
+    });
     return {
       content: [{
         type: "text",
         text: JSON.stringify({
+          view_url: viewUrl,
           samples_per_group: result.samplesEach,
           total_samples: result.samplesTotal,
           number_of_groups: params.num_variants + 1,
@@ -34977,10 +35028,19 @@ Example: "I have 50,000 total users and a 5% conversion rate. What's my MDE?"`,
     );
     const baselineValue = params.metric_type === "rate" ? params.baseline / 100 : params.baseline;
     const mdeRelative = baselineValue !== 0 ? mdeAbsolute / baselineValue : 0;
+    const viewUrl = buildDeeplink("/mde-calculator", {
+      t: String(params.total_traffic),
+      s: params.std_dev ? String(params.std_dev) : "",
+      r: String(params.baseline),
+      p: String(params.power),
+      c: String(params.confidence_level),
+      m: params.metric_type === "average" ? "c" : "r"
+    });
     return {
       content: [{
         type: "text",
         text: JSON.stringify({
+          view_url: viewUrl,
           mde_absolute: round(mdeAbsolute, 6),
           mde_relative_percent: toPercent(mdeRelative),
           interpretation: params.metric_type === "rate" ? `You can detect an absolute change of ${toPercent(mdeAbsolute)} (${toPercent(mdeRelative)} relative lift) from the ${params.baseline}% baseline.` : `You can detect an absolute change of ${round(mdeAbsolute, 4)} (${toPercent(mdeRelative)} relative lift) from the ${params.baseline} baseline.`,
@@ -35049,11 +35109,25 @@ Example: "Control: 5000 visitors, 250 conversions. Variant: 5000 visitors, 300 c
     }));
     const bestIdx = result.probabilityToBeBest.indexOf(Math.max(...result.probabilityToBeBest));
     const summary = result.probabilityToBeBest[bestIdx] > 0.95 ? `\u{1F3C6} ${params.variants[bestIdx].name} is the winner with ${toPercent(result.probabilityToBeBest[bestIdx])} probability of being best.` : result.probabilityToBeBest[bestIdx] > 0.8 ? `\u{1F4CA} ${params.variants[bestIdx].name} is leading with ${toPercent(result.probabilityToBeBest[bestIdx])} probability, but more data may be needed.` : `\u23F3 No clear winner yet. The leading variant has only ${toPercent(result.probabilityToBeBest[bestIdx])} probability of being best.`;
+    const isRate = params.metric_type === "rate";
+    const bayesPart = (v, withName) => {
+      const rate = isRate && v.conversions !== void 0 && v.users ? (v.conversions / v.users * 100).toFixed(2) : "";
+      return pipeJoin(isRate ? [v.users, v.conversions, rate, "", withName ? v.name : ""] : [v.users, v.mean, "", v.std_dev, withName ? v.name : ""]);
+    };
+    const viewUrl = buildDeeplink("/bayesian-calculator", {
+      c: bayesPart(params.variants[0], false),
+      v: params.variants.slice(1).map((v) => bayesPart(v, true)).join(";"),
+      m: isRate ? "r" : "a",
+      // alpha=beta=1 is the uninformative prior; anything else maps to "weak".
+      p: params.prior_alpha === 1 && params.prior_beta === 1 ? "u" : "w",
+      l: String(Math.round(params.credibility * 100))
+    });
     return {
       content: [{
         type: "text",
         text: JSON.stringify({
           summary,
+          view_url: viewUrl,
           variants: variantResults,
           lift_distribution: {
             mean: round(result.liftDistribution.mean, 6),
@@ -35095,11 +35169,16 @@ Example: "Control has 10,234 users, variant has 9,766 users"`,
     const total = params.control_users + params.variant_users;
     const actualRatio = params.control_users / total;
     const summary = result.hasMismatch ? `\u{1F6A8} SRM DETECTED (p=${round(result.pValue, 6)}) \u2014 The traffic split is ${toPercent(actualRatio)}/${toPercent(1 - actualRatio)} instead of the expected 50/50. Your experiment may have a bug. DO NOT trust the results.` : `\u2705 No SRM detected (p=${round(result.pValue, 4)}) \u2014 The traffic split of ${toPercent(actualRatio)}/${toPercent(1 - actualRatio)} is consistent with a 50/50 split.`;
+    const viewUrl = buildDeeplink("/sample-ratio-mismatch", {
+      c: String(params.control_users),
+      v: String(params.variant_users)
+    });
     return {
       content: [{
         type: "text",
         text: JSON.stringify({
           summary,
+          view_url: viewUrl,
           has_mismatch: result.hasMismatch,
           p_value: round(result.pValue, 6),
           chi_square: round(result.chiSquare, 4),
@@ -35191,10 +35270,19 @@ Example: "How many responses do I need from a population of 10,000 with 5% margi
       params.margin_of_error / 100,
       params.confidence_level
     );
+    const viewUrl = buildDeeplink("/sample-size-calculator", {
+      m: "r",
+      t: "s",
+      // survey mode
+      e: String(params.margin_of_error),
+      c: String(Math.round(params.confidence_level * 100)),
+      z: String(params.population)
+    });
     return {
       content: [{
         type: "text",
         text: JSON.stringify({
+          view_url: viewUrl,
           required_sample_size: sampleSize,
           population: params.population,
           margin_of_error: `\xB1${params.margin_of_error}%`,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "abtestresult-mcp",
-  "version": "1.0.8",
+  "version": "1.1.0",
   "description": "MCP server for A/B test statistical analysis — significance testing, sample size calculation, Bayesian analysis, and more",
   "license": "SEE LICENSE IN LICENSE",
   "type": "module",