@simpletoolsindiaorg/engi-mcp 1.2.0 → 1.2.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +8 -0
- package/README.md +75 -16
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -5,6 +5,14 @@ All notable changes to this project will be documented in this file.
|
|
|
5
5
|
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
|
|
6
6
|
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
|
|
7
7
|
|
|
8
|
+
## [1.2.1] - 2026-03-16
|
|
9
|
+
|
|
10
|
+
### Documentation
|
|
11
|
+
- Full benchmark results published: 15 scenarios across 2 real projects (engi-mcp repo + fresh sandbox), 51 files total
|
|
12
|
+
- Combined result: 225,983 → 8,716 tokens · **96.1% reduction**
|
|
13
|
+
- Added real API cost savings table (Claude Sonnet pricing · up to $418/month saved per developer at 50 tasks/day)
|
|
14
|
+
- Updated top-level banner and token-savings badge to reflect v1.2.0 numbers
|
|
15
|
+
|
|
8
16
|
## [1.2.0] - 2026-03-16
|
|
9
17
|
|
|
10
18
|
### Added
|
package/README.md
CHANGED
|
@@ -18,7 +18,7 @@
|
|
|
18
18
|
[](https://modelcontextprotocol.io)
|
|
19
19
|
[](LICENSE)
|
|
20
20
|
[](src/)
|
|
21
|
-
[](#-proven-token-savings)
|
|
22
22
|
|
|
23
23
|
</div>
|
|
24
24
|
|
|
@@ -31,10 +31,13 @@ Every time Claude Code helps you with a task, it reads files. Lots of them. For
|
|
|
31
31
|
**engi-mcp** gives Claude a compact intelligence layer: indexed summaries, scoped file discovery, and compact planning tools. Instead of reading 27 files (18,000 tokens), Claude reads a 350-token summary and gets the same job done.
|
|
32
32
|
|
|
33
33
|
```
|
|
34
|
-
WITHOUT engi-mcp: Claude reads
|
|
35
|
-
WITH engi-mcp: 3 tool calls →
|
|
34
|
+
WITHOUT engi-mcp: Claude reads 30 files → 20,842 tokens
|
|
35
|
+
WITH engi-mcp: 3 tool calls → 528 tokens
|
|
36
36
|
─────────────────────────
|
|
37
|
-
SAVED:
|
|
37
|
+
SAVED: 20,314 tokens (97.5% reduction)
|
|
38
|
+
|
|
39
|
+
Combined across 15 real scenarios · 51 files · 2 projects:
|
|
40
|
+
225,983 tokens → 8,716 tokens → 96.1% reduction
|
|
38
41
|
```
|
|
39
42
|
|
|
40
43
|
---
|
|
@@ -377,22 +380,78 @@ Resources are read via repo:// URIs — lighter than tools, no arguments needed.
|
|
|
377
380
|
|
|
378
381
|
## Proven Token Savings
|
|
379
382
|
|
|
380
|
-
|
|
383
|
+
Full benchmark run across **two real projects** — 15 scenarios, 51 files, every tool and resource exercised.
|
|
384
|
+
|
|
385
|
+
### Test 1 — engi-mcp own repo (30 files · 20,842 token baseline)
|
|
386
|
+
|
|
387
|
+
| Scenario | Tools | Without MCP | With MCP | Saved |
|
|
388
|
+
|----------|-------|------------|---------|-------|
|
|
389
|
+
| Analysis | 3 | 20,842 | 528 | **97.5%** |
|
|
390
|
+
| Bug Fix | 3 | 20,842 | 950 | **95.4%** |
|
|
391
|
+
| Feature | 5 | 20,842 | 1,159 | **94.4%** |
|
|
392
|
+
| POC | 2 | 20,842 | 128 | **99.4%** |
|
|
393
|
+
| Documentation | 3 | 20,842 | 443 | **97.9%** |
|
|
394
|
+
| Multi-session Memory | 2 | 41,684 | 422 | **99.0%** |
|
|
395
|
+
| All 7 Resources | 7 | 20,842 | 870 | **95.8%** |
|
|
396
|
+
| **TOTAL** | **19/19** | **166,736** | **4,500** | **97.3%** |
|
|
397
|
+
|
|
398
|
+
### Test 2 — Sandbox: fresh Node.js REST API (21 files · 6,583 token baseline)
|
|
399
|
+
|
|
400
|
+
> Project the MCP had never seen before — auth, users, db, notifications, utils.
|
|
401
|
+
|
|
402
|
+
| Scenario | Tools | Without MCP | With MCP | Saved |
|
|
403
|
+
|----------|-------|------------|---------|-------|
|
|
404
|
+
| Architecture Analysis | 3 | 6,583 | 194 | **97.1%** |
|
|
405
|
+
| Bug Fix | 3 | 6,583 | 866 | **86.8%** |
|
|
406
|
+
| Feature (OAuth2 login) | 5 | 6,583 | 854 | **87.0%** |
|
|
407
|
+
| Refactor (BaseRepo) | 4 | 6,583 | 813 | **87.7%** |
|
|
408
|
+
| POC (Redis rate limit) | 2 | 6,583 | 127 | **98.1%** |
|
|
409
|
+
| Documentation | 3 | 6,583 | 450 | **93.2%** |
|
|
410
|
+
| Multi-session Memory | 2 | 13,166 | 436 | **96.7%** |
|
|
411
|
+
| All 7 Resources | 7 | 6,583 | 476 | **92.8%** |
|
|
412
|
+
| **TOTAL** | **19/19** | **59,247** | **4,216** | **92.9%** |
|
|
413
|
+
|
|
414
|
+
### Combined across both tests
|
|
415
|
+
|
|
416
|
+
| | Tokens |
|
|
417
|
+
|--|--------|
|
|
418
|
+
| Without MCP | 225,983 |
|
|
419
|
+
| With MCP | 8,716 |
|
|
420
|
+
| **Saved** | **217,267** |
|
|
421
|
+
| **Net reduction** | **96.1%** |
|
|
381
422
|
|
|
382
|
-
|
|
383
|
-
|----------|-----------|------------|---------|-------|
|
|
384
|
-
| Analysis | 3 | 18,298 | 405 | **97.8%** |
|
|
385
|
-
| Bug Fix | 3 | 18,298 | 734 | **96.0%** |
|
|
386
|
-
| Feature | 5 | 18,298 | 1,117 | **93.9%** |
|
|
387
|
-
| POC | 2 | 18,298 | 128 | **99.3%** |
|
|
388
|
-
| Documentation | 3 | 18,298 | 226 | **98.8%** |
|
|
389
|
-
| Multi-session Memory | 2 | 36,596 | 434 | **98.8%** |
|
|
390
|
-
| All 7 Resources | 7 | 18,298 | 836 | **95.4%** |
|
|
391
|
-
| **TOTAL** | **19/19** | **146,384** | **3,880** | **97.3%** |
|
|
423
|
+
Total wall time: **103ms** for 15 scenarios. Tool coverage: **19/19 (100%)**.
|
|
392
424
|
|
|
393
|
-
|
|
425
|
+
### Real Cost Savings (Claude API pricing)
|
|
426
|
+
|
|
427
|
+
> Prices based on Claude Sonnet 4.5 input token rate ($3 / 1M tokens).
|
|
428
|
+
|
|
429
|
+
| Usage | Without MCP | With MCP | Saved / session |
|
|
430
|
+
|-------|------------|---------|----------------|
|
|
431
|
+
| 10 tasks/day · small repo (6k tok) | ~$0.18 | ~$0.013 | **$0.17** |
|
|
432
|
+
| 10 tasks/day · medium repo (21k tok) | ~$0.63 | ~$0.045 | **$0.58** |
|
|
433
|
+
| 10 tasks/day · large repo (100k tok) | ~$3.00 | ~$0.21 | **$2.79** |
|
|
434
|
+
| 50 tasks/day · large repo | ~$15.00 | ~$1.05 | **$13.95/day** |
|
|
435
|
+
|
|
436
|
+
At 50 tasks/day on a large repo: **~$418/month saved per developer.**
|
|
437
|
+
The larger the repo, the greater the savings — token cost scales linearly with repo size, MCP cost does not.
|
|
438
|
+
|
|
439
|
+
### What RAG adds (v1.2.0+)
|
|
440
|
+
|
|
441
|
+
engi-mcp v1.2.0 added a RAG engine that chunks file content and attaches real code snippets to summaries. The snippets eliminate follow-up file reads that the benchmark does not count:
|
|
442
|
+
|
|
443
|
+
| Tool | Extra tokens (RAG) | What it replaces |
|
|
444
|
+
|------|--------------------|-----------------|
|
|
445
|
+
| `flow_summarize` | +12 | 1–2 Read calls (~800 tok) |
|
|
446
|
+
| `bug_trace_compact` | +270 | 2–4 Read calls (~2,000 tok) |
|
|
447
|
+
| `doc_context_build` | +212 | 2–3 Read calls (~1,500 tok) |
|
|
448
|
+
|
|
449
|
+
Run the benchmarks yourself:
|
|
394
450
|
```bash
|
|
451
|
+
# Clone and run against this repo
|
|
395
452
|
npx tsx test-token-report.ts
|
|
453
|
+
|
|
454
|
+
# Or against your own project — edit REPO_PATH in test-token-report.ts
|
|
396
455
|
```
|
|
397
456
|
|
|
398
457
|
---
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@simpletoolsindiaorg/engi-mcp",
|
|
3
|
-
"version": "1.2.
|
|
3
|
+
"version": "1.2.1",
|
|
4
4
|
"description": "Software Engineering Intelligence MCP Server — 97% token reduction for Claude Code. 12 tools + 7 resources for scoped file discovery, flow summaries, bug tracing, planning, and multi-session memory.",
|
|
5
5
|
"main": "dist/index.js",
|
|
6
6
|
"types": "dist/index.d.ts",
|