claw-tsaver
An MCP server that helps AI agents reduce token usage by compressing, summarizing, and managing conversation/context data more efficiently.
README
claw-tsaver
A token-saving MCP proxy for OpenClaw users. Cuts tool call payloads by 90%+ via lazy expansion.
Why
MCP tool calls often return thousands of tokens of HTML or JSON in a single response — but the model typically uses only 5% of it. The remaining 95% silently burns context window and increases cost. claw-tsaver sits between OpenClaw and your downstream MCP servers, intercepts oversized responses, and hands the model a compact preview + an on-demand handle instead.
How
sequenceDiagram
participant U as OpenClaw (Claude)
participant C as claw-tsaver proxy
participant F as fetch / puppeteer / etc.
U->>C: call_tool("fetch", url)
C->>F: forward call
F-->>C: 11,507 tokens of HTML
Note over C: tiktoken count > threshold
C->>C: store full content in SQLite
C-->>U: {preview_head, preview_tail, expand_handle}<br/>(only 104 tokens)
Note over U: model decides if it needs full text
U->>C: expand_content(handle)
C-->>U: full 11,507 tokens
Real measurement
| Test | Original tokens | Returned tokens | Saved |
|---|---|---|---|
| fetch Wikipedia "Tokenization (data security)" | 11,507 | 104 | 99.1% |
Tested on OpenClaw + Claude Sonnet 4.6 + mcp-server-fetch, 2026-04-25.
Raw data: benchmarks/mvp-day1-fetch.jsonl.
Quick Start
1. Prerequisites
Install uv (one-time setup):
curl -LsSf https://astral.sh/uv/install.sh | sh
No claw-tsaver install needed — uvx will fetch and run it on demand.
2. Configure downstream MCP servers
Edit ~/.claw-tsaver/config.json (first run of claw-tsaver-mcp will auto-create a template):
{
"downstream_servers": [
{"name": "fetch", "command": "uvx", "args": ["mcp-server-fetch"]}
],
"compression_threshold_tokens": 500
}
3. Register with OpenClaw
Add this block at the top level of ~/.openclaw/openclaw.json:
"mcp": {
"servers": {
"claw-tsaver": {
"command": "uvx",
"args": ["--from", "git+https://github.com/Yang1Bai/claw-tsaver",
"claw-tsaver-mcp"]
}
}
}
Then restart OpenClaw gateway: openclaw gateway restart
Dashboard
Optional: a local web UI for real-time token savings stats.
uvx --from git+https://github.com/Yang1Bai/claw-tsaver claw-tsaver-dashboard
Open http://localhost:7878 in your browser.
Roadmap
- [x] Module A: lazy expansion proxy (this release)
- [x] Module D: local dashboard (this release)
- [ ] Module B: tool routing (auto-load only relevant MCPs per turn)
- [ ] Module C: conversation history compression (atomic fact cards)
License
MIT — see LICENSE file.
Contributing
Issues and PRs welcome.
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.
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.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.