claude-webcache
Cross-session WebFetch cache for Claude Code. Persists fetched docs across sessions via SQLite (7-day TTL) so the same URL+prompt in any new session is an instant cache hit instead of re-fetching
README
claude-webcache
Cross-session WebFetch cache for Claude Code.
Claude Code's built-in WebFetch caches results for 15 minutes within a single session. claude-webcache extends that across sessions, indefinitely (TTL 7 days, configurable).
Open new session -> your past fetches are still there.
Cache hit -> instant.
Cache miss -> same as built-in WebFetch.
Why
Any time you re-fetch the same URLs across sessions -- docs, API references, research pages -- you're paying the full fetch cost every time. The 15-minute in-session cache evicts before your next sprint. claude-webcache keeps fetches around so the second session hits cache instead.

Install
Option 1 -- Claude Code plugin (recommended)
One line, in your terminal:
claude plugin marketplace add theYahia/claude-webcache && claude plugin install claude-webcache@theyahia
Then add the usage pattern to your ~/.claude/CLAUDE.md (see Usage pattern).
💡 Why CLI subcommands and not the TUI
/plugin install? The TUI flow currently routes through Claude Code'sremoteMarketplaceClientbackend, which has an open server-side bug rejecting all third-party plugin sources withFailed to install: This plugin uses a source type your Claude Code version does not support.See anthropics/claude-code#41653 and ~20 related open issues. The CLI subcommands above useNativeMarketplaceReaderlocally and are unaffected.Fallback (if even CLI subcommands fail on your setup):
git clone https://github.com/theYahia/claude-webcache && claude --plugin-dir ./claude-webcache/pluginThis bypasses the marketplace pipeline entirely.
Option 2 -- npm global
npm i -g @theyahia/claude-webcache
Requires Node.js 22.5+ (uses built-in node:sqlite -- no native deps).
Then register manually in ~/.claude/settings.json (replace the path with the result of npm root -g):
{
"mcpServers": {
"claude-webcache": {
"command": "node",
"args": ["/path/from/npm-root-g/claude-webcache/scripts/mcp-server.cjs"]
}
},
"hooks": {
"SessionStart": [
{
"matcher": "startup|clear|compact",
"hooks": [
{ "type": "command", "command": "node /path/from/npm-root-g/claude-webcache/scripts/hook-stats.cjs" }
]
}
]
}
}
Option 3 -- clone (contributors)
See CONTRIBUTING.md.
Usage pattern
Add to your ~/.claude/CLAUDE.md so Claude routes web fetches through the cache:
## WebFetch caching (claude-webcache)
Before calling WebFetch, call `cached_fetch(url, prompt)` first.
- If it returns text -> use that, do NOT call WebFetch.
- If it returns `[CACHE_MISS] <url>` -> call WebFetch as normal, then call `cache_store(url, prompt, output)` with the result.
That's it. Same URL + same prompt in any future session = instant cache hit.
Tools (MCP)
| Tool | Args | Returns |
|---|---|---|
cached_fetch |
url, prompt |
cached text, or [CACHE_MISS] <url> |
cache_store |
url, prompt, output |
stored |
cache_stats |
-- | { total, hits, last } |
cache_list |
limit? |
recent URLs (most recent first) |
Storage
SQLite at ~/.webcache/cache.db (WAL mode, concurrent-safe). Cache key = SHA256(url + "|" + prompt).
| Field | Type |
|---|---|
key |
TEXT PRIMARY KEY |
url |
TEXT |
prompt_hash |
TEXT |
output |
TEXT |
cached_at |
INTEGER (ms epoch) |
hit_count |
INTEGER |
last_hit_at |
INTEGER |
SessionStart hook
On every new session, the hook injects a one-line stat:
webcache: 142 pages cached, 38 hits, last fetch 3h ago
Skips injection if cache is empty.
TTL
Default 7 days. Expired entries are deleted on next read of the same key. Run a manual purge by requiring src/cache.js and calling purgeExpired().
Limits
- Cache key includes the prompt -> different prompts on the same URL are separate entries. Pick consistent prompts (e.g. always
"extract title and main content") to maximize hit rate. - Output is whatever WebFetch returns (already summarized by the model). The cache doesn't re-process it.
- No semantic search, no embeddings. Exact
(url, prompt)match only.
License
MIT -- see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.