More screenshots
Works with the AI Agents you already use
SynRAG Token Saver: Local Code Search for Coding Agents
Make Claude Code, Cursor and other agents find the exact function instead of reading whole files: -63.9% tokens measured with a real agent.
Free
About this skill
Your coding agent reads whole files to find one function. Every one of those reads costs tokens.
This skill teaches your agent to use SynRAG, a free and open-source (MIT) MCP server that runs on your own machine, so it locates code first and reads only what it needs: the exact definition, who uses it and what breaks if you change it.
How SynRAG works (all local, CPU only, no paid APIs):
- Tree-sitter splits code by its syntax tree: one function, class, method or type per chunk, never cut in half
- BM25 search over LanceDB, reranked on CPU with FlashRank
- A dependency impact graph with resolved imports (tsconfig paths, barrels, monorepos and Python)
Measured with a real agent (Gemini 3.5 Flash-Lite, 10 multi-step coding tasks × 2 rounds, counting the tokens the API bills):
- -63.9% tokens vs working without SynRAG (95% CI: -83% to -24%)
- Correct answers: 87% → 94%
- 59-89 ms median per new query, on CPU
It also says where it loses: on very short tasks it spends more (+117%), because the tool definitions weigh more than the task. The skill tells your agent when not to use it, and the bajo_demanda profile cuts the fixed cost to 159 tokens per turn. Every number has a reproducible script in the repo (CLAIMS.md).
What's in the skill:
- The core loop: locate → read only the returned range → act
- 6 step-by-step playbooks: fix a bug, safe refactor, review a PR, onboard to a new repo, add a feature without duplicates, debug failing tests
- Reference for all 28 SynRAG tools with their parameters
- Setup guide (one-line installer, folders, tool profiles, verify, uninstall) and troubleshooting
Requirements: SynRAG installed locally (free, one command, Python 3.10+). Tested on Linux with Claude Code and Google Antigravity; the installer also configures Gemini CLI, Codex CLI, Cursor, Windsurf, Cline/Roo Code and Zed (not tested yet).
Engine and source: https://github.com/EnmanuelReynoso23/synrag
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 29 days ago
- Free with an Agensi account
Requires
Permissions
Allowed Hosts