ctxjev-mcp 0.1.11 → 0.2.0

package/README.md

@@ -8,7 +8,7 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 [![Node](https://img.shields.io/badge/node-%3E%3D20-339933?logo=node.js&logoColor=white)](package.json)
 
-[Setup](#setup) · [Tools](#tools) · [Example response](#example-response) · [Related packages](#related-packages)
+[Setup](#setup) · [Tools](#tools) · [Example Response](#example-response) · [Related Packages](#related-packages)
 
 </div>
 
@@ -17,6 +17,14 @@
 `ctxjev-mcp` speaks plain stdio MCP: the same binary works with every host below, and only the
 config shape differs.
 
+> **Before you expect it to save tokens:** an MCP tool returns data to whoever called it; it can't
+> remove anything from the host's own context. And to score its history, the agent has to send
+> that history as tool arguments, which the host model pays for in its own output tokens. It's
+> useful when something acts on the scores (an agent framework that manages its own context), not
+> as a drop-in token saver for Claude Code, Codex, or Copilot. For Claude Code, the
+> [ctxjev plugin](https://github.com/x96x64/ctxjev#the-claude-code-plugin) is the integration that
+> actually helps.
+
 ## Setup
 
 Get a key at [console.typesafe.ai/settings/keys](https://console.typesafe.ai/settings/keys) (no
@@ -59,7 +67,11 @@ codex mcp add ctxjev --env TYPESAFE_API_KEY=... -- npx ctxjev-mcp
 
 `entries` is `{ id, role: 'user'|'assistant'|'tool', toolName?, content, timestamp }[]`.
 
-## Example response
+Entry content and the goal are sent to TypeSafe AI's Jev API. Common secret formats (API keys,
+tokens, private-key blocks, `NAME=value` credentials) are masked to `[REDACTED]` first, on a
+best-effort basis.
+
+## Example Response
 
 Calling `prune_history` with two entries, one obviously relevant to the goal and one not, returns:
 
@@ -71,7 +83,7 @@ Calling `prune_history` with two entries, one obviously relevant to the goal and
   ],
   "savings": {
     "totalEntries": 2, "keptEntries": 1, "droppedEntries": 1, "summarizedEntries": 0,
-    "totalTokens": 13, "savedTokens": 5
+    "totalTokens": 13, "droppedTokens": 5, "summarizableTokens": 0
   },
   "usage": { "inputTokens": 406, "outputTokens": 36 }
 }
@@ -79,7 +91,7 @@ Calling `prune_history` with two entries, one obviously relevant to the goal and
 
 This is a real response body, captured against the live API.
 
-## Related packages
+## Related Packages
 
 | Package | What it is |
 | --- | --- |
@@ -91,6 +103,7 @@ Full docs, design notes, and per-host setup live in the main repo:
 
 ## License
 
-[MIT](LICENSE): free to use, modify, and distribute, including in a commercial product, as long
-as the license text and copyright notice ship with it. See the [main repo](https://github.com/x96x64/ctxjev#license)
-for how this matches every dependency `ctxjev` currently uses.
+This package is released under the [MIT](LICENSE) license: free to use, modify, and distribute,
+including in a commercial product, as long as the license text and copyright notice ship with it.
+See the [main repo](https://github.com/x96x64/ctxjev#license) for how this matches every
+dependency `ctxjev` currently uses.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ctxjev-mcp",
-  "version": "0.1.11",
+  "version": "0.2.0",
   "description": "MCP server exposing ctxjev's Jev-based context scoring/pruning as tools for any MCP-capable agent host.",
   "keywords": [
     "mcp",
@@ -40,7 +40,7 @@
   "dependencies": {
     "@modelcontextprotocol/sdk": "^1.30.0",
     "zod": "^4.6.5",
-    "ctxjev-core": "0.1.11"
+    "ctxjev-core": "0.2.0"
   },
   "devDependencies": {
     "@types/node": "^22",