Vibe Monitor
Vibe Monitor
A local architecture advisor for AI coding workflows. AI agents ship features fast; Vibe Monitor helps you see when architectural complexity is accumulating and what actually needs attention.
The open-source MCP server
This repository contains an actual local MCP server (stdio transport), not only an npm installer or a cloud dashboard. The MCP server runs on your own computer and exposes read-only architecture checks and change-history snapshots to an MCP-capable coding agent. It does not rewrite your project source.
Example MCP tools implemented in src/mcp.js:
MCP tool | What it does locally |
| Start or resume a bounded assessment of one coding project |
| Read the current project's architecture signals without creating a snapshot |
| Record a post-change snapshot to compare structural evolution |
| Show prioritized findings and concrete next steps |
| Inspect architecture evolution across recorded checkpoints |
| Inspect the actual running/installed version |
Install and register the local MCP server with supported agents:
npx -y --registry=https://registry.npmjs.org @vibe-monitor/setup@latest installDirect stdio MCP command (for clients that configure command / args themselves; Node.js 22+):
npx -y --registry=https://registry.npmjs.org @vibe-monitor/setup@latest mcpThe npm package implements both the installer and MCP server. The server's official name is io.github.tubban1/vibe-monitor; its registry manifest declares the local mcp argument. Some analysis features depend on the signed Cloud scoring policy; installing does not grant cloud access to your source code. See the telemetry details below.
Related MCP server: AI Knowledge Center MCP
Get started — official stable release
The official open-source MCP server and installer are distributed through npm as @vibe-monitor/setup@latest. As verified on 2026-11-03, npm latest is @vibe-monitor/setup@1.0.14; the GitHub release and npm release workflow are complete.
npx -y --registry=https://registry.npmjs.org @vibe-monitor/setup@latest install
On a new install, bounded pseudonymous product telemetry defaults ON, with notice. Use install --no-telemetry or vibe-monitor telemetry off to disable collection; this pauses Free architecture analysis. An existing opt-out survives reinstall. No activation code is required.
Not sure whether your projects need it? Give your coding Agent this one canonical instruction:
Analyze accessible code projects and return a direct recommendation →
The Agent must inspect code projects already accessible in its current workspaces, decide which would benefit (if any), and report the results to you in the chat. No installation or external report submission occurs during the assessment.
Your source code is not uploaded
Vibe Monitor analyzes source locally. It never uploads source files, diffs, prompts, project paths or secrets to Vibe Monitor servers. Bounded derived metrics and product-operation events can be sent through disclosed pseudonymous telemetry, not the code itself. You can audit the open-source implementation instead of relying on a marketing promise:
The coding Agent itself (Codex, Claude Code, Cursor, etc.) may process code under its own provider's terms. That is separate from Vibe Monitor's own upload behavior.
What you gain
Evidence before alerts: architecture pressure, coupling, module responsibilities and tests.
Practical priorities: what to improve now, what to leave alone and when to reassess.
Historical perspective: how architectural signals evolve across coding sessions.
Low overhead: local structural scans do not call an LLM; optional Agent-assisted interpretation may use tokens.
Vibe Monitor scans project code read-only. Installation registers supported Agent MCP/rules configuration and may create local history/backups. No separately installed always-on monitoring daemon is required.
Documentation
Official installation and verification · Agent evaluation/report protocol · Illustrative architectural scenarios · Security · Telemetry/privacy · Website
Open source
The unified client is Apache-2.0 licensed and publicly auditable here. Signed Cloud policy/scoring configuration is a separate proprietary service; policy delivery does not grant permission to upload raw source or execute arbitrary remote code. Official release provenance: GitHub Releases and npm @vibe-monitor/setup.
Available Tools
19 toolsvibe_account_connectA
Explain how to link this Vibe Monitor installation to the user's account. Call ONLY when the human asks to link this device. Makes no network request.
Inputs: project_path.
vibe_checkpointA
Checkpoint structural evolution after a meaningful implementation batch. Persists a snapshot and reports stage-aware findings.
Inputs: path.
vibe_evolution_auditC
Read-only source-based structural duplicate candidates, legacy consumer-reference review, and multi-snapshot code evolution with Git churn; no automatic deletion or causal claims.
Inputs: path.
vibe_explain_findingC
Explain one finding with evidence, standards mapping, heuristic disclosure, staged action and over-engineering guard.
Inputs: id (required), path.
vibe_feedbackA
Record explicit user feedback about a Vibe Monitor finding. Use only when the user clearly says a warning was useful, wrong, too early, too late, or confusing. Never ask for feedback after every warning.
Inputs: id (required), verdict (required), note, path.
vibe_finding_intelligenceB
Read risk trend, evidence-review status and explicitly confirmed Product Intent guidance for one Finding. Includes an unverified baseline for future feedback, never causal attribution.
Inputs: id (required), path.
vibe_findingsB
Return current findings ordered by evolution stage and pressure score. Findings include Do now, Prepare, Do not do yet, and Next trigger.
Inputs: path.
vibe_get_architecture_review_requestB
Create a bounded ARP v1 semantic architecture review request for a finding that explicitly recommends semantic review. The request contains normalized architecture evidence and no raw source code or project path.
Inputs: id (required), path, independentReview.
vibe_legacy_auditA
Find local legacy compatibility, retired feature markers, disabled branches and possibly unreferenced private functions. Each candidate requires consumer verification; nothing is auto-deleted.
Inputs: path.
vibe_outcome_record_actionA
ONLY after a user or agent explicitly states they applied, deferred, or dismissed a specific Vibe recommendation. Record a REPORTED action, not inferred code changes; never claim a fix occurred. Idempotent via stable submissionId.
Inputs: id (required), actor (required), disposition (required), path, +2 more.
vibe_outcome_statusB
Read local recommendation→reported action→comparable follow-up→user verification and recurrence history. No project scan or Cloud telemetry. Risk improvement is observational, attribution remains unproven.
Inputs: id, path, limit.
vibe_outcome_verifyA
Record ONLY explicit user-reported verification for a previously reported action. User affirmation plus regression-test/independent-review evidence type is still not proof of causation. Never fabricate execution or user confirmation.
Inputs: actor (required), verdict (required), outcomeId (required), evidenceType (required), +1 more.
vibe_product_intentA
Read Vibe Monitor's current product-intent context. Inferred values are advisory; explicit values come only from the user's own stated intent.
Inputs: path.
vibe_project_statusA
Read current architecture-evolution status without creating a history snapshot.
Inputs: path.
vibe_session_startB
Start or resume Vibe Monitor for the current coding session. Can be interrupted or explicitly skipped without preventing coding. For an explicit user Skip request, set mode=skip. Otherwise establishes a bounded project assessment.
Inputs: mode, path.
vibe_set_product_intentB
Store explicit product-intent facts locally when the user has clearly stated them. Do not turn guesses into explicit intent and do not ask these questions unless an architecture decision materially depends on them.
Inputs: path, scale, tenancy, audience, +3 more.
vibe_skip_currentA
Skip an in-progress Vibe Monitor Session Start, at the user's explicit request. Cancels active scanning in this MCP server without stopping coding or disabling future monitoring. Agent hosts may expose this tool as an action; Vibe Monitor cannot insert custom buttons in host UI.
Inputs: no named parameters.
vibe_submit_architecture_reviewB
Submit one ARP v1 semantic architecture review. Different agents may legitimately return different verdicts. Full model-generated review text stays local; Cloud receives only bounded structured review metadata for comparison and outcome learning.
Inputs: verdict (required), reviewId (required), confidence (required), recommendedStage (required), +8 more.
vibe_versionA
Get the Vibe Monitor version currently running in this MCP process, the installed current/previous runtime versions, staged update, and update state. A process may still run an older bundle after the installed pointer changes. Use this tool when the user asks which Vibe Monitor version is running or installed. schemaVersion and evidenceVersion are data format versions, NOT the software version.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 19 tools
Several tools have overlapping read-oriented purposes: vibe_project_status, vibe_findings, vibe_finding_intelligence and vibe_evolution_audit all surface status/risk information, and vibe_explain_finding vs vibe_finding_intelligence both target a single finding. The verbose descriptions do help differentiate (intelligence = trend/evidence, explain = evidence/standards/action), and the vibe_outcome_* trio is cleanly separated, but the boundaries still require careful reading.
The vibe_ prefix and snake_case are used consistently throughout, which is good. However, the ordering convention is mixed: some names are verb_noun (vibe_explain_finding, vibe_set_product_intent, vibe_skip_current) while others are noun-first (vibe_session_start, vibe_project_status, vibe_outcome_record_action, vibe_checkpoint, vibe_findings). Readable but not predictable.
At 19 tools this sits in the heavy-but-arguably-justified band for a monitoring/audit product with distinct lifecycle phases (session, audit, findings, feedback, outcome, architecture review). Each tool does map to a real concept, but the surface is large enough that several near-synonymous read tools could likely be consolidated.
The recommendation lifecycle is unusually well covered end-to-end: findings, explanation, feedback, reported action, verification, and outcome status, plus session start/skip, checkpoint, version, and architecture review request/submit. Minor gaps exist (no tool to suppress/dismiss or list findings history/sessions, no account unlink), but agents can work around these.
Maintenance
Related MCP Connectors
Remote MCP server exposing your Node Command fleet (files, shell, processes, browser) to AI hosts.
Artifact store for AI agents. Hosted OAuth at mcp.artifacta.io/mcp; local stdio via npm/PyPI.
Shared memory MCP server for Claude Code, Codex, Gemini CLI and Cursor. Cited, editable.
Repository knowledge graph MCP server for codebase understanding and debugging.
Related MCP Servers
- AlicenseBqualityDmaintenanceLocal MCP server providing project cognition capabilities for AI coding agents, including context packs, impact analysis, and git diff review through stdio communication.103MIT
- AlicenseNot gradedqualityCmaintenanceLocal-first MCP server that provides project context, verification gates, and structured tools for coding agents to discover knowledge, run diagnostics, and execute allowlisted commands within a repository.41 npmMIT
- AlicenseNot gradedqualityBmaintenanceMCP server that studies a folder of code repositories, generates markdown documentation (architecture, interfaces, activity), and serves answers over MCP with live ripgrep search and file reads so AI assistants can query codebases and their relationships.29 npmMIT
- AlicenseAqualityBmaintenanceLocal MCP server for AI agents and vibe coding safety: deterministic risk checks (payments, auth, database, secrets, infrastructure), sessions, checkpoints, policy inspection, and fix prompts. Runs over stdio against a local git repository; no language model judges risk.8Apache 2.0