hound-mcp 2.3.0__tar.gz

hound_mcp-2.3.0/.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,21 @@
+---
+name: Bug report
+about: Something isn't working
+title: "[Bug] "
+labels: bug
+assignees: dondai1234
+---
+
+**What happened**
+<!-- What did you try to do? What went wrong? -->
+
+**URL or query**
+<!-- The URL you tried to fetch, or the search query you used -->
+
+**Error message**
+<!-- Paste the full error if available -->
+
+**Environment**
+- Python version:
+- OS (Windows / macOS / Linux):
+- Hound version: (`hound --version` or check pyproject.toml)

hound_mcp-2.3.0/.github/workflows/ci.yml

@@ -0,0 +1,34 @@
+name: CI
+
+on:
+  push:
+    branches: [master]
+  pull_request:
+    branches: [master]
+
+jobs:
+  test:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, windows-latest]
+        python-version: ["3.11", "3.12"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e .[dev]
+        shell: bash
+
+      - name: Run tests
+        run: python -m pytest tests/ -v --tb=short
+        shell: bash

hound_mcp-2.3.0/.gitignore

@@ -0,0 +1,30 @@
+# Python
+__pycache__/
+*.py[cod]
+*.so
+*.egg-info/
+dist/
+build/
+*.egg
+
+# Virtual environment
+.venv/
+venv/
+env/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Project
+.master_fetch_cache/
+*.log
+
+# Test artifacts
+TEST/exa_himeno.txt

hound_mcp-2.3.0/CHANGELOG.md

@@ -0,0 +1,92 @@
+# Changelog
+
+## [2.3.0] - 2026-06-01
+
+### Fixed
+- **Smart router now detects JS-only shells and escalates** (#1): When HTTP returns a 200 with content like "You need to enable JavaScript", smart_fetch now escalates to dynamic. When dynamic returns a JS-disabled placeholder (e.g. Twitter), escalates to stealthy. Previously accepted these as successful results.
+- **Error field now signals content quality issues** (#7, #8): The `error` field is set when content is a JS shell (`js_shell_detected`), a geo/region redirect (`geo_redirect_detected`), or a bot challenge page (`bot_challenge_detected`). Downstream agents can now detect failures without parsing content strings.
+- **Bulk output now respects max_content_chars** (#3): All bulk operations (`bulk_get`, `bulk_fetch`, `bulk_stealthy_fetch`) now accept a `max_content_chars` parameter (default 40000) that truncates each result. Prevents 300KB+ output that overwhelms tool runtimes.
+- **Bulk `successful` count now excludes results with content issues**: A result with an error field is no longer counted as successful.
+
+### Changed
+- **Domain intelligence expanded**: Added known-safe domains (httpbin.org, wikipedia.org, github.com, stackoverflow.com, etc.) to prevent over-escalation of static sites. Added YouTube, Uniswap, Spotify, Notion, and other SPA domains as known-dynamic. Moved Twitter/X from dynamic to stealthy (dynamic returns JS-disabled placeholder for these).
+- **All smart_fetch return paths now run content quality annotation**: Every exit point (Phase A/B/C, force_fetcher, escalation results) calls `_annotate_quality()` to ensure the error field is populated when content is bad.
+- **All-tiers-failed error now includes failure trace**: The `error` field shows which tiers were tried and what failed.
+
+### Added
+- 15 new unit tests covering JS shell detection, content quality, geo redirect, domain intelligence routing
+
+## [2.0.0] - 2026-06-02: Hound
+
+Renamed product from "Master Fetch" to **Hound**: web research for AI agents.
+Internal module name stays `master_fetch`. Package: `hound-mcp`. CLI: `hound`.
+
+### Added
+
+### Added
+- Web search via TinyFish API: `smart_search` tool returns structured results (title, url, snippet)
+  - Free (30 searches/min), no API key needed
+  - Results cached for 5 minutes via SQLite
+  - Optional install: `pip install master-fetch[all]`
+  - Fetch-only users stay lean with zero extra dependencies
+
+### Changed
+- Package architecture: `master-fetch` = fetch only, `master-fetch[all]` = fetch + search
+- README rewritten with competitor comparison tables and one-prompt install guides
+
+## [1.1.0] - 2026-06-02
+
+### Added
+- Robots.txt compliance: respects site scraping policies by default. `respect_robots=False` to bypass.
+- HTTP retry logic: exponential backoff (1s/2s/4s) on transient network failures
+- Comprehensive test suite: 22 unit tests covering models, chunking, CF detection, domain extraction, binary detection, robots.txt
+- GitHub Actions CI: cross-platform testing on Ubuntu and Windows (Python 3.11, 3.12)
+- Proper PyPI metadata: classifiers, dev dependencies, pytest config
+
+### Changed
+- ResponseModel now includes `extracted_type`, `session_id`, `duration_ms`, `error` fields
+- Error messages: all-tiers-failed returns diagnostic trace showing what was tried
+
+### Fixed
+- Binary content (PDF) no longer crashes the extractor, returns clean error
+- HTTP error pages (non-challenge) no longer trigger wasteful browser escalation
+
+## [1.0.4] - 2026-06-02
+
+### Fixed
+- PDF and binary content handling: returns clean error instead of crashing with decode error
+- HTTP error pages no longer trigger unnecessary browser escalation. If the response contains no bot challenge text, the error is returned directly.
+
+## [1.0.3] - 2026-06-02
+
+### Fixed
+- Domain extraction now correctly handles multi-part TLDs (.co.uk, .com.au, .co.jp, etc.)
+- Auto-persistent browser sessions in smart_fetch. Dynamic and stealthy tiers now reuse browser instances instead of launching a new browser each time.
+- Disabled resource loading in dynamic/stealthy tiers for ~25% speed improvement.
+- Rewrote README with accurate, minimal information.
+
+## [1.0.2] - 2026-06-01
+
+### Fixed
+- Default caching was OFF (cache_ttl=0). Now defaults to 3600s (1 hour). Repeat fetches return instantly from cache.
+- Added 40KB content chunking with offset continuation. AI agents get a truncation notice when content exceeds the limit.
+
+## [1.0.0] - 2026-06-01
+
+### Added
+- Smart fetch routing: auto-escalates HTTP → Dynamic → Stealthy based on bot detection
+- Cloudflare Turnstile/Interstitial bypass via Patchright + Scrapling
+- Trafilatura content extraction pipeline (markdown, text, article, structured)
+- SQLite content caching with configurable TTL
+- Domain intelligence system: remembers which domains need which fetcher level
+- 12 MCP tools: get, bulk_get, fetch, bulk_fetch, stealthy_fetch, bulk_stealthy_fetch, screenshot, open_session, close_session, list_sessions, smart_fetch, cache_clear
+- Streamable HTTP transport (--http flag) for remote agent connections
+- Anti-bot bypass for DataDome, Akamai, Cloudflare challenges
+- Content quality rating: 9.5/10 vs competitors
+- Beats Exa and Tavily on JS-rendered and bot-protected pages (see COMPARISON_REPORT.md)
+
+### Known Limitations
+- DataDome + Cloudflare dual protection (g2.com) still blocks all fetchers
+- Reddit infinite scroll only returns first-load content
+- No built-in rate limiting between fetcher tiers
+- Domain extraction doesn't handle .co.uk / .com.au correctly

hound_mcp-2.3.0/CONTRIBUTING.md

@@ -0,0 +1,36 @@
+# Contributing
+
+Issues and pull requests are welcome. AI-generated issues are fine. Just make sure they're real problems you encountered while using Hound, not hypotheticals.
+
+## Reporting issues
+
+Open an issue on GitHub. Include:
+
+- What you were trying to do
+- The URL you tried to fetch or query you searched
+- The error or unexpected behavior
+- Your Python version and OS
+
+Bug report and feature request templates are available when you open a new issue.
+
+## Pull requests
+
+- Keep changes focused. One problem per PR.
+- Run `pytest tests/` before submitting.
+- If adding features, include tests.
+
+## Development
+
+```bash
+git clone https://github.com/dondai1234/master-fetch.git
+cd master-fetch
+pip install -e .[all,dev]
+playwright install chromium
+pytest tests/
+```
+
+The fetch engine is in `src/master_fetch/server.py`. Search is in `src/master_fetch/search.py`.
+
+## License
+
+MIT. By contributing, you agree to license your work under the same terms.

hound_mcp-2.3.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Bishesh Bhandari
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

hound_mcp-2.3.0/PKG-INFO

@@ -0,0 +1,28 @@
+Metadata-Version: 2.4
+Name: hound-mcp
+Version: 2.3.0
+Summary: Web research for AI agents. Fetch any page with anti-bot bypass plus web search. $0 forever.
+Project-URL: Repository, https://github.com/dondai1234/master-fetch
+Project-URL: Issues, https://github.com/dondai1234/master-fetch/issues
+Project-URL: Changelog, https://github.com/dondai1234/master-fetch/blob/master/CHANGELOG.md
+Author-email: Bishesh Bhandari <bishesh@master-fetch.dev>
+License-File: LICENSE
+Keywords: ai-agent,anti-bot,cloudflare-bypass,content-extraction,mcp,mcp-server,web-scraping
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Framework :: AsyncIO
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Internet :: WWW/HTTP :: Dynamic Content
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Requires-Python: >=3.11
+Requires-Dist: aiosqlite>=0.20.0
+Requires-Dist: scrapling[ai]>=0.4.7
+Requires-Dist: trafilatura>=2.0.0
+Provides-Extra: all
+Requires-Dist: requests>=2.31; extra == 'all'
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0; extra == 'dev'

hound_mcp-2.3.0/README.md

@@ -0,0 +1,177 @@
+# Hound
+
+<p align="center">
+  <img src="https://img.shields.io/badge/cost-$0%20forever-brightgreen" alt="$0 forever">
+  <img src="https://img.shields.io/badge/bypass-Cloudflare-blue" alt="Cloudflare bypass">
+  <img src="https://img.shields.io/badge/mcp-stdio-purple" alt="MCP">
+  <img src="https://img.shields.io/github/license/dondai1234/master-fetch" alt="MIT">
+</p>
+
+<p align="center">
+  <strong>Web research for AI agents. Fetch any page. Search the web. $0 forever.</strong>
+</p>
+
+<p align="center">
+  Bypasses Cloudflare bot protection. Extracts clean content. Built-in web search.<br>
+  No API keys needed. No accounts. No Docker. One MCP server.
+</p>
+
+> The fetch engine is called Master Fetch internally (`master_fetch` module). The product is Hound. Repo: `dondai1234/master-fetch`.
+
+---
+
+## Tools
+
+| Tool | Does |
+|------|------|
+| `smart_fetch` | Fetches any URL. Auto-escalates if blocked. Returns clean markdown. |
+| `smart_search` | Searches the web. Returns title, URL, snippet. Free, 30/min. |
+| `fetch` / `stealthy_fetch` | Manual tier selection. |
+| `screenshot` | Full-page screenshot of any URL. |
+| `open_session` / `close_session` | Persistent browser sessions. |
+| `bulk_fetch` / `bulk_stealthy_fetch` | Parallel fetching. |
+| `cache_clear` | Clear the cache. |
+
+---
+
+## Install via AI agent
+
+Paste the prompt below. The agent does the work. You just provide an API key when asked.
+
+### Full package (fetch + search)
+
+```
+Install Hound MCP server on this machine. Steps:
+
+1. Run: pip install hound-mcp[all]
+2. Run: playwright install chromium
+3. Add Hound to the MCP config on this machine. The server name is "hound". The command is "hound" with no arguments. Create a backup of the config file before editing.
+4. After setup, guide me to get a free TinyFish API key:
+
+   Tell me: "Go to https://tinyfish.ai, click Get API Key, sign up (free, no credit card needed). Copy the key and paste it here."
+
+   Wait for me to paste the key. Do not proceed until I provide it.
+5. Once I give you the key, add it as an environment variable TINYFISH_API_KEY in the hound server's MCP config.
+6. Tell me: "Done. Restart your agent to apply. You should see smart_fetch and smart_search in your tools."
+```
+
+### Fetch only
+
+```
+Install Hound MCP server (fetch only) on this machine. Steps:
+
+1. Run: pip install hound-mcp
+2. Run: playwright install chromium
+3. Add Hound to the MCP config on this machine. The server name is "hound". The command is "hound" with no arguments. Create a backup of the config file before editing.
+4. Tell me: "Done. Restart your agent to apply. You should see smart_fetch in your tools."
+```
+
+---
+
+## Updating
+
+```bash
+pip install --upgrade hound-mcp[all]
+```
+
+Then restart your agent. MCP servers launch fresh on each session, so the new version is picked up automatically. No config changes needed.
+
+---
+
+## Manual install
+
+```bash
+# Full package
+git clone https://github.com/dondai1234/master-fetch.git
+cd master-fetch
+pip install -e .[all]
+playwright install chromium
+
+# Fetch only
+git clone https://github.com/dondai1234/master-fetch.git
+cd master-fetch
+pip install -e .
+playwright install chromium
+```
+
+Add to MCP config (`mcpServers` / `mcp.servers`):
+
+```json
+{
+  "mcpServers": {
+    "hound": {
+      "command": "hound",
+      "env": {
+        "TINYFISH_API_KEY": "sk-tinyfish-..."
+      }
+    }
+  }
+}
+```
+
+A built-in API key is included and works out of the box. Get your own free key at [tinyfish.ai](https://tinyfish.ai) for higher limits.
+
+---
+
+## How fetching works
+
+`smart_fetch` tries the fastest method first, escalates only if blocked:
+
+| Tier | Engine | Speed | Best for |
+|------|--------|-------|----------|
+| HTTP | curl_cffi (Chrome TLS) | 1-3s | Most websites |
+| Dynamic | Playwright/Chromium | 3-8s | JS-heavy pages |
+| Stealthy | Patchright + solver | 5-13s | Pages behind bot protection |
+
+It remembers which tier works per domain. Results are cached (SQLite, 1hr TTL) so repeat fetches are instant.
+
+---
+
+## Fetch comparison
+
+Hound's fetch engine (Master Fetch) vs every alternative. Tested live, June 2026.
+
+| | Hound | Exa | Tavily | Crawl4AI | Jina |
+|---|---|---|---|---|---|
+| **Cloudflare bypass** | ✅ Built-in | ❌ | ❌ | ⚠️ Needs external proxies | ❌ |
+| **Auto-escalation** | ✅ HTTP→Browser→Stealth | ❌ | ❌ | ⚠️ Proxy rotation | ❌ |
+| **Domain intelligence** | ✅ Remembers per-domain | ❌ | ❌ | ❌ | ❌ |
+| **Content caching** | ✅ SQLite, instant hits | ❌ | ❌ | ❌ | ❌ |
+| **Persistent sessions** | ✅ 2x speed on repeats | ❌ | ❌ | ❌ | ❌ |
+| **Retry logic** | ✅ Exponential backoff | ❌ | ❌ | ✅ Via proxies | ❌ |
+| **Runs on your hardware** | ✅ | ❌ | ❌ | ✅ | ❌ |
+| **Cost** | $0 forever | 1K/mo free | 1K/mo free | $0 | Free tier |
+| **MCP native** | ✅ Stdio | ❌ | ❌ | ❌ | ❌ |
+
+**Real-world example:** Fetching a Cloudflare-protected page:
+- Hound: escalates to stealthy, solves challenge, returns full content (5-13s)
+- Exa/Tavily: HTTP 403. No fallback. Returns error.
+- Crawl4AI: 403 unless you bring your own proxy service. Returns error or empty.
+- Jina: 403. Returns error.
+
+## Full package comparison
+
+| | Hound | Exa | Tavily |
+|---|---|---|---|
+| **Web search** | ✅ 30/min | ✅ | ✅ |
+| **Content fetching** | ✅ + anti-bot bypass | ✅ | ✅ |
+| **All-in-one MCP** | ✅ One server | ❌ Two APIs | ❌ Two APIs |
+| **Cost** | $0 forever | 1K/mo free | 1K/mo free |
+
+---
+
+## Limits
+
+- DataDome, Akamai Bot Manager, and Cloudflare Turnstile (interactive): not bypassed. No free tool can bypass these. Hound is the closest you can get at $0. The next step up is paid proxy services or enterprise scraping APIs.
+- Reddit: stealthy tier, first page only (no infinite scroll)
+- YouTube: minimal text (expected for video pages)
+
+## Requirements
+
+- Python 3.11+
+- Chromium: `playwright install chromium`
+- Search: `pip install hound-mcp[all]`
+
+## License
+
+MIT

hound_mcp-2.3.0/pyproject.toml

@@ -0,0 +1,50 @@
+[project]
+name = "hound-mcp"
+version = "2.3.0"
+description = "Web research for AI agents. Fetch any page with anti-bot bypass plus web search. $0 forever."
+authors = [{name = "Bishesh Bhandari", email = "bishesh@master-fetch.dev"}]
+requires-python = ">=3.11"
+dependencies = [
+    "scrapling[ai]>=0.4.7",
+    "trafilatura>=2.0.0",
+    "aiosqlite>=0.20.0",
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Internet :: WWW/HTTP :: Dynamic Content",
+    "Topic :: Software Development :: Libraries :: Python Modules",
+    "Intended Audience :: Developers",
+    "Environment :: Console",
+    "Framework :: AsyncIO",
+]
+keywords = ["mcp", "web-scraping", "cloudflare-bypass", "anti-bot", "content-extraction", "ai-agent", "mcp-server"]
+
+[project.optional-dependencies]
+dev = [
+    "pytest>=8.0",
+]
+all = [
+    "requests>=2.31",
+]
+
+[project.urls]
+Repository = "https://github.com/dondai1234/master-fetch"
+Issues = "https://github.com/dondai1234/master-fetch/issues"
+Changelog = "https://github.com/dondai1234/master-fetch/blob/master/CHANGELOG.md"
+
+[project.scripts]
+hound = "master_fetch.server:main"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/master_fetch"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]

hound_mcp-2.3.0/src/master_fetch/__init__.py

@@ -0,0 +1 @@
+"""Master Fetch MCP Server."""

hound_mcp-2.3.0/src/master_fetch/cache.py

@@ -0,0 +1,115 @@
+"""SQLite-based content cache with TTL support.
+
+Stores fetched content keyed by URL+params hash. Auto-expires entries past TTL.
+"""
+import hashlib
+import json
+import time
+from pathlib import Path
+
+import aiosqlite
+
+# Default cache dir: next to the project
+_CACHE_DIR = Path.home() / ".master_fetch_cache"
+_DB_NAME = "cache.db"
+
+DEFAULT_TTL = 3600  # 1 hour
+
+
+def _cache_key(url: str, extraction_type: str, css_selector: str | None = None) -> str:
+    """Deterministic cache key from fetch params."""
+    raw = f"{url}|{extraction_type}|{css_selector or ''}"
+    return hashlib.sha256(raw.encode()).hexdigest()[:24]
+
+
+async def _ensure_db(cache_dir: Path | None = None) -> Path:
+    """Ensure the DB and table exist. Returns DB path."""
+    d = cache_dir or _CACHE_DIR
+    d.mkdir(parents=True, exist_ok=True)
+    db_path = d / _DB_NAME
+
+    async with aiosqlite.connect(db_path) as db:
+        await db.execute("""
+            CREATE TABLE IF NOT EXISTS cache (
+                key TEXT PRIMARY KEY,
+                url TEXT NOT NULL,
+                extraction_type TEXT NOT NULL,
+                content TEXT NOT NULL,
+                status INTEGER NOT NULL,
+                fetched_at REAL NOT NULL,
+                ttl INTEGER NOT NULL DEFAULT 3600
+            )
+        """)
+        await db.execute("CREATE INDEX IF NOT EXISTS idx_fetched_at ON cache(fetched_at)")
+        await db.commit()
+
+    return db_path
+
+
+async def get_cached(
+    url: str,
+    extraction_type: str,
+    css_selector: str | None = None,
+    ttl: int = DEFAULT_TTL,
+    cache_dir: Path | None = None,
+) -> dict | None:
+    """Return cached response if fresh, else None."""
+    key = _cache_key(url, extraction_type, css_selector)
+    db_path = await _ensure_db(cache_dir)
+
+    async with aiosqlite.connect(db_path) as db:
+        db.row_factory = aiosqlite.Row
+        cursor = await db.execute(
+            "SELECT * FROM cache WHERE key = ? AND fetched_at + ttl > ?",
+            (key, time.time()),
+        )
+        row = await cursor.fetchone()
+        if row is None:
+            return None
+        return {
+            "status": row["status"],
+            "content": json.loads(row["content"]),
+            "url": row["url"],
+        }
+
+
+async def set_cached(
+    url: str,
+    extraction_type: str,
+    content: list[str],
+    status: int,
+    css_selector: str | None = None,
+    ttl: int = DEFAULT_TTL,
+    cache_dir: Path | None = None,
+) -> None:
+    """Store a response in cache."""
+    key = _cache_key(url, extraction_type, css_selector)
+    db_path = await _ensure_db(cache_dir)
+
+    async with aiosqlite.connect(db_path) as db:
+        await db.execute(
+            """INSERT OR REPLACE INTO cache (key, url, extraction_type, content, status, fetched_at, ttl)
+               VALUES (?, ?, ?, ?, ?, ?, ?)""",
+            (key, url, extraction_type, json.dumps(content), status, time.time(), ttl),
+        )
+        await db.commit()
+
+
+async def clear_cache(cache_dir: Path | None = None) -> int:
+    """Clear all expired entries. Returns count of purged rows."""
+    db_path = await _ensure_db(cache_dir)
+    async with aiosqlite.connect(db_path) as db:
+        cursor = await db.execute(
+            "DELETE FROM cache WHERE fetched_at + ttl <= ?", (time.time(),)
+        )
+        await db.commit()
+        return cursor.rowcount
+
+
+async def clear_all_cache(cache_dir: Path | None = None) -> int:
+    """Nuke the entire cache. Returns count of purged rows."""
+    db_path = await _ensure_db(cache_dir)
+    async with aiosqlite.connect(db_path) as db:
+        cursor = await db.execute("DELETE FROM cache")
+        await db.commit()
+        return cursor.rowcount