Add TOKEN_SAVINGS.md to explain jCodeMunch MCP

Document the benefits and savings of using jCodeMunch MCP for token management in AI agents.

Author: jgravelle

TOKEN_SAVINGS.md

@@ -0,0 +1,55 @@
+# jcodemunch-mcp — TOKEN_SAVINGS.md
+
+# Token Savings: jCodeMunch MCP
+
+## Why This Exists
+
+AI agents waste tokens when they must read entire files to locate a single function, class, or constant.
+jCodeMunch indexes a repository once and allows agents to retrieve **exact symbols on demand**, eliminating unnecessary context loading.
+
+---
+
+## Example Scenario
+
+**Repository:** Medium Python codebase (300+ files)
+**Task:** Locate and read the `authenticate()` implementation
+
+| Approach         | Tokens Consumed | Process                               |
+| ---------------- | --------------- | ------------------------------------- |
+| Raw file loading | ~7,500 tokens   | Open multiple files and scan manually |
+| jCodeMunch MCP   | ~1,449 tokens   | `search_symbols` → `get_symbol`       |
+
+**Savings:** ~80.7%
+
+---
+
+## Typical Savings by Task
+
+| Task                     | Raw Approach | With jCodeMunch | Savings |
+| ------------------------ | ------------ | --------------- | ------- |
+| Explore repo structure   | ~200k tokens | ~2k tokens      | ~99%    |
+| Find a specific function | ~40k tokens  | ~200 tokens     | ~99.5%  |
+| Read one implementation  | ~40k tokens  | ~500 tokens     | ~98.7%  |
+| Understand module API    | ~15k tokens  | ~800 tokens     | ~94.7%  |
+
+---
+
+## Scaling Impact
+
+| Queries | Raw Tokens | Indexed Tokens | Savings |
+| ------- | ---------- | -------------- | ------- |
+| 10      | 400k       | ~5k            | 98.7%   |
+| 100     | 4M         | ~50k           | 98.7%   |
+| 1,000   | 40M        | ~500k          | 98.7%   |
+
+---
+
+## Key Insight
+
+jCodeMunch shifts the workflow from:
+
+**“Read everything to find something”**
+to
+**“Find something, then read only that.”**
+
+---