@cachly-dev/mcp-server 0.10.65 → 0.10.66

package/dist/src/handlers/brain.js

@@ -6,7 +6,7 @@ import { keywordSearch, tokenize, splitMultiQuery, levenshtein, indexVocab as _i
 import { rerankByQuality } from '../rerank.js';
 import { computeEmbedding, hasEmbedProvider } from '../embeddings.js';
 // ── Changelog (shown once per version in session_start) ──────────────────────
-const MCP_VERSION = '0.10.65';
+const MCP_VERSION = '0.10.66';
 const WHATS_NEW = {
     '0.10.64': [
         `📈 **The three decisive metrics, now measurable**`,

package/dist/src/index.js

@@ -64,7 +64,7 @@ import { Redis } from 'ioredis';
 const API_URL = process.env.CACHLY_API_URL ?? 'https://api.cachly.dev';
 let JWT = process.env.CACHLY_JWT ?? '';
 const _EMBED_MODEL = process.env.CACHLY_EMBED_MODEL ?? '';
-const CURRENT_VERSION = '0.10.65';
+const CURRENT_VERSION = '0.10.66';
 // Max time to wait for a freshly-provisioned instance to become "running" before
 // giving up. Free-tier provisioning in high-latency regions can take 45–90s, so the
 // old hard-coded 25s caused spurious "instance_not_reachable" on first run.

package/package.json

@@ -1,9 +1,15 @@
 {
   "name": "@cachly-dev/mcp-server",
-  "version": "0.10.65",
+  "version": "0.10.66",
   "mcpName": "io.github.cachly-dev/cachly-mcp",
   "description": "Your AI forgets everything between sessions. cachly fixes that permanently. Persistent memory + causal root-cause analysis for Claude Code, Cursor, Copilot & Windsurf. Learns from every bug fix, git commit, and CI run automatically. Arrives pre-briefed. Predicts failures before they happen. 101 MCP tools · Free tier forever · GDPR · EU servers.",
   "type": "module",
+  "files": [
+    "dist/src/**/*",
+    "dist/packages/**/*",
+    "scripts/postinstall.js",
+    "server.json"
+  ],
   "bin": {
     "cachly-mcp": "dist/src/index.js"
   },

package/server.json

@@ -6,13 +6,13 @@
     "url": "https://github.com/cachly-dev/cachly-mcp",
     "source": "github"
   },
-  "version": "0.10.65",
+  "version": "0.10.66",
   "packages": [
     {
       "registryType": "npm",
       "name": "@cachly-dev/mcp-server",
       "identifier": "@cachly-dev/mcp-server",
-      "version": "0.10.65",
+      "version": "0.10.66",
       "transport": {
         "type": "stdio"
       }

package/.env.test.example

@@ -1,23 +0,0 @@
-# E2E test environment variables — copy to .env.test and fill in values.
-# Run: cp .env.test.example .env.test && npx vitest run src/__tests__/e2e/
-
-# API base URL (default: production API)
-E2E_API_URL=https://api.cachly.dev
-
-# Valid Keycloak JWT or long-lived API key (cky_live_...)
-# Get from: cachly.dev/instances → Settings → API Token
-E2E_JWT=
-
-# UUID of an active (running) Brain instance for this token
-E2E_INSTANCE_ID=
-
-# Redis password for the test instance (from connection details)
-E2E_API_KEY=
-
-# Redis host and port for the test instance
-E2E_REDIS_HOST=
-E2E_REDIS_PORT=6380
-E2E_REDIS_TLS=true
-
-# Set to "true" to skip slow tests (provisioning, device flow)
-E2E_SKIP_SLOW=false

package/.github/ISSUE_TEMPLATE/bug_report.md

@@ -1,50 +0,0 @@
----
-name: Bug report
-about: Something isn't working as expected
-title: '[bug] '
-labels: bug
-assignees: ''
----
-
-## What happened?
-
-<!-- Describe the bug clearly. -->
-
-## Steps to reproduce
-
-1.
-2.
-3.
-
-## Expected behavior
-
-<!-- What should have happened? -->
-
-## Brain doctor output
-
-Run `brain_doctor` in your AI assistant and paste the output here.
-
-<details>
-<summary>brain_doctor output</summary>
-
-```
-(paste here)
-```
-</details>
-
-## Environment
-
-- Package version: <!-- `npm list @cachly-dev/mcp-server` -->
-- Editor / AI assistant: <!-- Claude Code / Cursor / Windsurf / etc. -->
-- OS: <!-- macOS / Linux / Windows -->
-- Node version: <!-- `node --version` -->
-
-## Logs
-
-<details>
-<summary>Relevant logs or error output</summary>
-
-```
-(paste here)
-```
-</details>

package/.github/ISSUE_TEMPLATE/config.yml

@@ -1,8 +0,0 @@
-blank_issues_enabled: false
-contact_links:
-  - name: Documentation
-    url: https://cachly.dev/docs/mcp
-    about: Read the full MCP server docs
-  - name: Discord / Community
-    url: https://cachly.dev/discord
-    about: Ask questions and share what you've built

package/.github/ISSUE_TEMPLATE/feature_request.md

@@ -1,25 +0,0 @@
----
-name: Feature request
-about: Suggest a new tool, behavior, or improvement
-title: '[feat] '
-labels: enhancement
-assignees: ''
----
-
-## What problem does this solve?
-
-<!-- Describe the pain point clearly. What can't you do today that you wish you could? -->
-
-## Proposed solution
-
-<!-- How would you like it to work? Be specific — tool name, input/output, behavior. -->
-
-## Alternatives you've considered
-
-<!-- What workarounds are you using now? What else did you consider? -->
-
-## Context
-
-- Tool(s) involved: <!-- e.g. smart_recall, causal_trace, brain_predict -->
-- Editor / AI assistant: <!-- Claude Code / Cursor / Windsurf / etc. -->
-- Use case: <!-- team / solo / CI / other -->

package/.github/workflows/eas-build-android.yml

@@ -1,58 +0,0 @@
-name: EAS Build — TCO Android Preview
-
-on:
-  workflow_dispatch:
-    inputs:
-      message:
-        description: 'Build reason'
-        default: 'manual build'
-
-jobs:
-  build:
-    name: Build Android preview APK
-    runs-on: ubuntu-latest
-    defaults:
-      run:
-        working-directory: apps/travel-chaos-organizer/frontend
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: '20'
-
-      - name: Install dependencies
-        run: npm install
-
-      - name: Convert SVG icons to PNG (EAS Build requires PNG)
-        run: |
-          sudo apt-get install -y --no-install-recommends librsvg2-bin
-          rsvg-convert -w 1024 -h 1024 assets/icon.svg        -o assets/icon.png
-          rsvg-convert -w 1024 -h 1024 assets/adaptive-icon.svg -o assets/adaptive-icon.png
-          rsvg-convert -w 2048 -h 2048 assets/splash.svg        -o assets/splash.png
-          echo "Icons converted:"
-          ls -lh assets/*.png
-
-      - name: Setup EAS
-        uses: expo/expo-github-action@v8
-        with:
-          eas-version: latest
-          token: ${{ secrets.EXPO_TOKEN }}
-
-      - name: Create .env with production variables
-        run: |
-          echo "EXPO_PUBLIC_API_URL=${{ secrets.EXPO_PUBLIC_API_URL }}" >> .env
-          echo "EXPO_PUBLIC_KEYCLOAK_URL=${{ secrets.EXPO_PUBLIC_KEYCLOAK_URL }}" >> .env
-          echo "EXPO_PUBLIC_KEYCLOAK_REALM=${{ secrets.EXPO_PUBLIC_KEYCLOAK_REALM || 'travel-chaos' }}" >> .env
-          echo "EXPO_PUBLIC_KEYCLOAK_CLIENT_ID=${{ secrets.EXPO_PUBLIC_KEYCLOAK_CLIENT_ID || 'tco-mobile' }}" >> .env
-
-      - name: Build Android (preview — internal distribution)
-        run: |
-          eas build \
-            --profile preview \
-            --platform android \
-            --non-interactive \
-            --no-wait
-        # --no-wait: EAS build runs in the cloud asynchronously.
-        # Download the APK from expo.dev → your project → Builds when done (~10-15 min).

package/.github/workflows/eas-update.yml

@@ -1,55 +0,0 @@
-name: EAS Update — TCO Production
-
-on:
-  push:
-    branches: [main]
-    paths:
-      - 'apps/travel-chaos-organizer/frontend/**'
-  workflow_dispatch:
-    inputs:
-      message:
-        description: 'Update message'
-        default: 'OTA update'
-
-jobs:
-  update:
-    name: Publish EAS Update to production
-    runs-on: ubuntu-latest
-    defaults:
-      run:
-        working-directory: apps/travel-chaos-organizer/frontend
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: '20'
-
-      - name: Install dependencies
-        run: npm ci
-
-      - name: TypeScript typecheck
-        run: npx tsc --noEmit
-
-      - name: Create .env with production variables
-        run: |
-          echo "EXPO_PUBLIC_API_URL=${{ secrets.EXPO_PUBLIC_API_URL }}" >> .env
-          echo "EXPO_PUBLIC_KEYCLOAK_URL=${{ secrets.EXPO_PUBLIC_KEYCLOAK_URL }}" >> .env
-          echo "EXPO_PUBLIC_KEYCLOAK_REALM=${{ secrets.EXPO_PUBLIC_KEYCLOAK_REALM || 'travel-chaos' }}" >> .env
-          echo "EXPO_PUBLIC_KEYCLOAK_CLIENT_ID=${{ secrets.EXPO_PUBLIC_KEYCLOAK_CLIENT_ID || 'tco-mobile' }}" >> .env
-
-      - name: Publish EAS Update
-        env:
-          EXPO_TOKEN: ${{ secrets.EXPO_TOKEN }}
-        run: |
-          MESSAGE="${{ github.event.inputs.message || github.event.head_commit.message }}"
-          MESSAGE="${MESSAGE:0:100}"
-          # Publish to both channels so preview and production builds both get the update
-          npx eas-cli update \
-            --branch production \
-            --message "$MESSAGE" \
-            --non-interactive
-          npx eas-cli update \
-            --branch preview \
-            --message "$MESSAGE" \
-            --non-interactive

package/.github/workflows/publish.yml

@@ -1,72 +0,0 @@
-name: Publish to npm
-
-on:
-  push:
-    branches: [main]
-  workflow_dispatch:
-
-jobs:
-  publish:
-    runs-on: ubuntu-latest
-    permissions:
-      contents: write  # needed to create GitHub Releases
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: '20'
-          registry-url: 'https://registry.npmjs.org'
-
-      - name: Install dependencies
-        run: npm install --ignore-scripts
-
-      - name: Test
-        run: npm test
-
-      - name: Build
-        run: npm run build
-
-      - name: Check if version is already published
-        id: version_check
-        run: |
-          PKG_VERSION=$(node -p "require('./package.json').version")
-          echo "version=$PKG_VERSION" >> $GITHUB_OUTPUT
-          if npm view "@cachly-dev/mcp-server@$PKG_VERSION" version &>/dev/null; then
-            echo "skip=true" >> $GITHUB_OUTPUT
-            echo "Version $PKG_VERSION already published — skipping."
-          else
-            echo "skip=false" >> $GITHUB_OUTPUT
-            echo "Version $PKG_VERSION is new — will publish."
-          fi
-
-      - name: Extract CHANGELOG notes
-        if: steps.version_check.outputs.skip == 'false'
-        id: changelog
-        run: |
-          VERSION="${{ steps.version_check.outputs.version }}"
-          # Extract section for this version from CHANGELOG.md
-          NOTES=$(awk "/^## \[?${VERSION}\]?/{found=1; next} found && /^## /{exit} found{print}" CHANGELOG.md | sed '/^[[:space:]]*$/d' | head -60)
-          if [ -z "$NOTES" ]; then
-            NOTES="See [CHANGELOG.md](https://github.com/cachly-dev/cachly-mcp/blob/main/CHANGELOG.md) for details."
-          fi
-          # Write to file to avoid env var quoting issues
-          echo "$NOTES" > /tmp/release-notes.md
-          echo "See the [CHANGELOG](https://github.com/cachly-dev/cachly-mcp/blob/main/CHANGELOG.md) for full details." >> /tmp/release-notes.md
-
-      - name: Publish
-        if: steps.version_check.outputs.skip == 'false'
-        run: npm publish --access public
-        env:
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
-
-      - name: Create GitHub Release
-        if: steps.version_check.outputs.skip == 'false'
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          VERSION="${{ steps.version_check.outputs.version }}"
-          gh release create "v${VERSION}" \
-            --title "v${VERSION}" \
-            --notes-file /tmp/release-notes.md \
-            --latest

package/BENCH.md

@@ -1,113 +0,0 @@
-# Cachly-Bench — the recall-quality proof
-
-> The STRATEGY.md moat claim is: *Cachly's structured, quality-aware recall beats
-> raw text matching (and a flat-file memory) at surfacing the lesson that actually
-> helps.* A claim without a number is marketing. This is the number.
-
-## Run it
-
-```bash
-npm run bench            # pretty report
-npm run bench -- --json  # machine-readable
-```
-
-The benchmark is also a CI regression guard: `src/__tests__/rerank.test.ts`
-asserts that quality reranking never regresses the headline metrics and that the
-baseline is already strong (so the corpus isn't rigged).
-
-## What it measures
-
-Two rankers are compared over the **same** BM25 candidate set, so the comparison
-isolates *ranking quality*, not retrieval:
-
-| Ranker | What it does | Analogy |
-|---|---|---|
-| `flatfile` | naive term-overlap, no IDF/quality, recency tiebreak | an LLM reading `/memories` files |
-| `baseline` | raw BM25+ keyword score (`src/search.ts`) | a decent text search engine |
-| `cachly` | BM25+ then quality-aware rerank (`src/rerank.ts`) | structured, trust-aware recall |
-
-Quality reranking boosts a lesson by five orthogonal signals:
-1. **Outcome** (proven success > partial > failed attempt)
-2. **Confidence** (0..1 → maps to a 0.7..1.3 nudge)
-3. **Proven-ness** (`recall_count`, log-saturated, caps at +30%)
-4. **Severity** (critical issue > minor, mild tiebreaker)
-5. **Governance** (`review_level` × distinct `endorsements` — a senior-reviewed
-   lesson outranks an unreviewed one even when the text is nearly identical)
-
-All five factors are multiplied and clamped to `[0.5, 2.5]` so a strong text match
-is never fully buried, and a weak one never catapulted on quality alone.
-
-## The corpus is fair (not rigged)
-
-The fixture corpus (`src/bench/fixtures.ts`) is deliberately and *fairly*
-adversarial in two dimensions:
-
-**Failure-distractor pairs:** for several queries a **symptom-dense failed attempt**
-competes with the **solution-focused proven success**. Failed attempts repeat the
-symptom verbatim ("still stuck, still failing"); the lesson that fixed it is written
-in solution vocabulary. BM25 alone is fooled by the symptom density; quality
-reranking corrects it.
-
-**Governance pairs:** two success lessons with near-identical vocabulary, but only
-one has a senior reviewer endorsement. BM25 cannot distinguish them; the governance
-signal breaks the tie in favour of the reviewed, canonical lesson.
-
-The baseline still scores well (MRR ~82%, Precision@1 ~69%) — we are not making a
-weak baseline look bad. We are improving an already-good one on the cases that matter.
-
-## Current results
-
-Three rankers, same corpus:
-
-- **flat-file** — naive term-overlap, no IDF, no length normalization, **no quality
-  signal**, ties broken by recency. This is an honest (in fact *charitable*) stand-in
-  for "an LLM reading its own `/memories` files": it has no ranking engine.
-- **baseline** — raw BM25+ keyword ranking.
-- **cachly** — BM25+ followed by quality-aware reranking (`rerank.ts`).
-
-```
-  corpus: 17 lessons · queries: 13
-
-  metric       flatfile baseline   cachly    vs flat
-  Precision@1     76.9%    69.2%    84.6%     +10.0%
-  Precision@3     37.2%    33.3%    33.3%     -10.3%
-  Recall@3       100.0%    96.2%    96.2%      -3.8%
-  MRR             87.2%    82.1%    91.0%      +4.4%
-  nDCG@5          89.9%    85.5%    92.4%      +2.8%
-
-  vs BM25 baseline : MRR +10.9% · Precision@1 +22.2%
-  vs flat-file mem : MRR  +4.4% · Precision@1 +10.0%
-```
-
-**Headline: Precision@1 +22.2% vs BM25, +10.0% vs a flat-file memory.**
-
-Interpretation — read the metrics that matter for an *agent*: it acts on the **top**
-result, so **Precision@1** and **MRR** are decisive. There, cachly wins against both
-baselines: it puts the single proven fix first most often. On the hardest cases — a
-symptom-dense failed attempt that beats the success on raw text, or two near-identical
-success lessons where only one was reviewed — quality reranking surfaces the lesson
-that actually solves the problem.
-
-**Why is flat-file's Precision@3 / Recall@3 higher?** Honesty matters: the flat-file
-ranker scores the *entire* 17-lesson corpus, while cachly and BM25 rank the capped
-BM25 candidate set. On a tiny corpus, "return more" mechanically inflates recall-at-k
-without making the *top* answer better. We deliberately did **not** rig this away — and
-we **steelman** the flat-file with a recency tiebreak it wouldn't normally have. Even
-so, cachly wins the metrics an agent actually depends on.
-
-## Honest limitations (read this)
-
-- **Small corpus.** 17 lessons, 13 queries. This proves the *mechanism* works for
-  each quality signal and guards against regressions. It does **not** yet prove
-  real-world magnitude. The flat-file head-to-head is now built in (a naive,
-  quality-blind ranker), but the next step is a larger, externally-sourced corpus
-  and the same head-to-head on **real agent traces** with third-party labels.
-- **Ranking, not retrieval.** Both rankers see the same BM25 candidates, so this
-  measures the quality layer only. CKG-traversal and semantic retrieval are
-  evaluated separately via integration tests and are not in these numbers.
-- **Labels are ours.** Gold relevance is author-assigned. A credible public
-  benchmark needs third-party labels.
-
-The point of this file is not to claim victory. It is to make the moat
-*measurable* and *defended by CI*, so every future change to recall must move this
-number in the right direction — or it doesn't ship.