logslim-mcp
Compacts noisy test, build, and cloud-log output before an AI agent reads it — dedupes repeats, collapses stack frames, folds Playwright retries, and renders CloudWatch/GCP JSON logs down to the signal. Typically 80–95% fewer tokens on failures. Tool: compact_output. Run: npx -y logslim logslim-mcp
README
logslim
CI failed? Get a 5-line PR summary — not a 400-line Actions log.
Agent reading test output? Cut 80–95% of the tokens.
When tests or builds fail, you scroll GitHub Actions logs. When Claude Code or Cursor runs
npm test, the agent reads everything — progress bars, 120 identical warnings, 40 frames
of node_modules. logslim fixes both:
- CI / humans — GitHub Action posts structured failures on your PR (file, line, fix hints)
- Agents / tokens — CLI + MCP compacts noisy output before an LLM reads it (~80–95% savings on failures)
npx logslim -- npm test
No account. No API key. MIT open source.
GitHub Action — PR failure summary
When CI fails, post a readable summary on the pull request instead of making reviewers dig through Actions logs.
permissions:
contents: read
pull-requests: write
jobs:
test:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: 20
- run: npm ci
- name: Run tests
id: test
run: npm test 2>&1 | tee test-output.log
continue-on-error: true
- name: Post failure summary
if: steps.test.outcome == 'failure' && github.event_name == 'pull_request'
uses: P156HAM/logslim/action@v0.4.0
with:
log-file: test-output.log
exit-code: 1
github-token: ${{ secrets.GITHUB_TOKEN }}
- name: Fail job
if: steps.test.outcome == 'failure'
run: exit 1
What gets posted on the PR:
- Structured failures with
file:lineand messages - Fix hints for known codes (
TS2339,ERESOLVE, …) - Link to the full CI log
- Token/log reduction stats (useful when agents also read the output)
The Action uses the same engine as the CLI — compaction, error extraction, and code cards. You get human-readable PR comments and agent-ready JSON from one tool.
| Input | Default | Purpose |
|---|---|---|
log-file |
(required) | Path to captured test/build log |
exit-code |
1 |
Exit code of the failed command |
github-token |
(required) | secrets.GITHUB_TOKEN with pull-requests: write |
logslim-version |
0.3.0 |
npm version to run |
skip-on-success |
true |
Don't comment when no failure detected |
See it work (30 seconds)
git clone https://github.com/P156HAM/logslim.git
cd logslim && npm install && npm run build && npm run demo
Or without cloning:
npx logslim -- node -e "console.log('ok'); for(let i=0;i<30;i++)console.log('warn '+i); throw Error('fail')"
Before → After
BEFORE — what the agent reads today (~18 lines here; real runs are 500–3000):
PASS src/utils.test.ts
console.warn deprecated prop id=1000
console.warn deprecated prop id=1001
console.warn deprecated prop id=1002
... (same warning ×120)
FAIL src/checkout/cart.test.ts
Expected: 89.10
Received: 99.00
at cart.test.ts:48:27
at node_modules/jest-circus/build/utils.js:298:28
at node_modules/jest-circus/build/utils.js:231:10
at node_modules/jest-circus/build/run.js:252:3
... (15 more node_modules frames)
Test Suites: 1 failed, 1 passed, 2 total
AFTER — what logslim gives the agent:
PASS src/utils.test.ts
console.warn deprecated prop id=1000
console.warn deprecated prop id=1001
console.warn deprecated prop id=1002
(+5 similar lines omitted by logslim)
FAIL src/checkout/cart.test.ts
Expected: 89.10
Received: 99.00
at cart.test.ts:48:27
at node_modules/jest-circus/build/utils.js:298:28
… 3 vendor/internal frames collapsed by logslim
Test Suites: 1 failed, 1 passed, 2 total
Same failure. Same fix. ~92% fewer tokens.
How it works
npm test ──► logslim ──► agent / CI / you
│
├─ 1. Strip ANSI colors & spinner garbage
├─ 2. Dedupe repeated lines (warn spam)
├─ 3. Collapse node_modules stack frames
├─ 4. Group similar lines (timestamps/ids masked)
├─ 5. Extract structured errors (file, line, message)
├─ 6. Attach fix cards for known codes (TS2339, ERESOLVE…)
└─ 7. Optional token budget (trim middle, keep errors)
Failure mode (default): tests pass → light cleanup only. Tests fail → full pipeline. You only pay the compaction cost when something actually broke.
Every removed section is marked in place ((+47 similar lines omitted by logslim))
so the agent knows data was elided and can re-run the raw command if needed.
Install
npm install -g logslim
# or zero-install:
npx logslim -- npm test
Requires Node 18+.
Usage
Basic — wrap any command
logslim -- npm test
logslim -- python -m pytest -x
logslim -- npx tsc --noEmit
Exit code is preserved. Output on stdout is compacted. Stats on stderr.
Pipe mode
npm test 2>&1 | logslim
npm test; logslim --exit-code $? 2>&1 < full.log # if you saved output
JSON — for agents and CI
logslim --json -- npm test
{
"exitCode": 1,
"failed": true,
"compacted": "FAIL src/checkout/cart.test.ts\n...",
"errors": [
{
"file": "cart.test.ts",
"line": 48,
"message": "Expected: 89.10, Received: 99.00",
"kind": "assertion"
}
],
"codes": [
{
"id": "TS2339",
"lang": "typescript",
"meaning": "Property does not exist on type",
"fix_steps": [
"Check for typos",
"Extend the interface",
"Use optional chaining"
]
}
],
"stats": {
"tokensIn": 3296,
"tokensOut": 252,
"saved": 0.92,
"applied": "full"
}
}
The agent reads compacted + errors + codes — not thousands of lines of prose.
CI context (GitHub Actions)
logslim --json --attach git,ci -- npm test
Prepends: branch: feat/x | commit: a3f2c1d | pr: #42 (from GITHUB_* env vars).
All options
| Flag | What it does |
|---|---|
--mode failure |
Compact hard only on failure (default) |
--mode full |
Always compact hard |
--mode light |
Strip ANSI only, never aggressive dedupe |
--json |
Structured output (see above) |
--attach git,ci |
Prepend branch/commit/CI metadata |
--budget 2000 |
Hard token cap; errors + head/tail survive |
--exit-code N |
For pipe mode when you know the exit code |
--no-codes |
Skip error code fix cards |
--no-stats |
Hide stderr savings footer |
MCP server (Claude Code / Cursor)
Lets the agent call compaction as a tool — no manual piping.
Project .mcp.json or Claude Desktop config:
{
"mcpServers": {
"logslim": {
"command": "npx",
"args": ["-y", "logslim-mcp"]
}
}
}
Tool: compact_output — pass output (raw log text) and optional exit_code.
Returns compacted text, extracted errors, fix cards, and stats.
Local dev:
npm run build && npm run mcp
Tell your agent to use it
Add to CLAUDE.md, AGENTS.md, or .cursor/rules:
When running tests, builds, or linters that produce verbose output:
- Prefer: `logslim --mode failure --json -- <command>`
- Read the `compacted`, `errors`, and `codes` fields before debugging.
- If output was elided, re-run the raw command only if you need full logs.
Error code fix cards
When logs contain known codes, logslim attaches a short fix card (~30 tokens) instead of making the agent guess or search docs.
| Family | Examples | Source |
|---|---|---|
| TypeScript | TS2339, TS2554, TS2307 | errors/typescript.json |
| Node | ECONNREFUSED, ENOTFOUND | errors/node.json |
| npm | ERESOLVE, ELIFECYCLE | errors/npm.json |
Hand-curated pocket references — not scraped docs. PRs welcome to add codes.
Measured savings
| Log type | Lines | Tokens | Saved |
|---|---|---|---|
| Jest (warn spam + 1 failure) | 149 → 25 | ~3,300 → ~250 | 92% |
| Webpack build (asset noise + 2 TS errors) | 548 → 55 | ~8,900 → ~1,000 | 88% |
| Pytest (25 identical failures) | 356 → 153 | ~4,300 → ~1,500 | 64% |
Token counts are estimated (~4 chars/token). Good for relative savings, not billing.
Library API
import { compact, process } from "logslim";
const { text, stats } = compact(rawLog, { mode: "failure", exitCode: 1 });
const result = process(rawLog, {
mode: "failure",
exitCode: 1,
attach: ["git", "ci"],
});
// result.text, result.errors, result.codes, result.stats
When to use logslim
| Use it | Skip it |
|---|---|
| CI failed and you want a PR summary, not a 400-line log | Tests passed and output is already short |
| AI agents running tests/builds locally or in CI | You already tee full logs to disk for audit |
| Long repetitive failure output (jest, pytest, webpack) | Platform already truncates well enough for you |
| MCP workflows where tool output hits context limits | You need full logs for compliance archive |
Keep full logs if you need them:
npm test 2>&1 | tee full.log | logslim
Development
npm install
npm test
npm run build
npm run demo
Contributing
logslim gets sharper every time it learns a new error code or a new log format — and both are easy first contributions:
- Add an error fix card (TypeScript / Node / npm) — a ~5-minute, pure-JSON PR.
- Share a log that compacts badly — paste real output from a tool logslim mangles.
- Add support for a new runner — Playwright, pytest, vitest, cargo, gradle…
Start here: good first issues · CONTRIBUTING.md
License
MIT — use freely, no account required.
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.