MCP Server · stdio transport
AI assistants
type fast.
They don't engineer.
nablr (pronounced like "enabler") connects to Cursor, Claude Desktop, or Windsurf and enforces quality gates across your entire software delivery lifecycle.
$ uvx nablr
// connects nablr as an MCP server
> "Start my nablr trial"
✓ 14-day Pro trial activated, no card required
2-minute setup · no credit card · disable anytime
The problem it solves
AI assistants accelerate typing, not engineering.
72–120hr → 8.0hr
Source: SEF: A Multi-Agent Orchestration Framework for Parallel SDLC Execution, DOI 10.5281/zenodo.18410037
Who reaches for nablr
Illustrative profiles, not customer quotes: the roles this is actually built for.
Needs board-level visibility into platform health before a reorg, acquisition, or hiring push.
Inherits a codebase with no docs and no safe entry point for change.
Ships fast with Cursor or Claude Code, needs a governance layer without hiring a team.
One orchestrated pipeline
18 expert agents run as a structured handoff sequence; each validates the prior phase before the next begins.
Plus 7 specialist agents on call: Forensic Engineer Strategist Platform Engineer Impact Architect Migration Planner Parity Tester Cutover Engineer
Nine entry points, one pipeline
Each starts the same orchestrated sequence above: pick the one matching where you are today.
Six patterns, one MCP server
Governance proves a step happened. It doesn't automatically prove the content was correct. nablr does both, and here's exactly how.
18 expert personas, not one blanket instruction: a security lens, a requirements lens, an architecture lens.
Work can't silently skip a stage: a requirement gets validated before development starts, and a change can't merge until it clears defined criteria.
8 scanner engines, 40+ rules, run independently of whatever the AI assistant claims it did. They complement human review; they don't replace it.
Low-risk fixes resolve on their own; higher-risk ones always stop and wait for a human decision before they ship.
Hand-offs, decisions, and escalations get recorded, so you can reconstruct what happened later.
Problems spotted mid-task get recorded rather than silently fixed out of scope, or silently ignored.
Free vs Pro
Every scanner works free. Persona expertise is the Pro layer.
Pick your plan
Annual saves the most. Monthly keeps it flexible.
One outbound domain. That's it.
nablr makes exactly one category of outbound call: license validation. Your code never leaves your machine.
The only domain nablr ever calls: license checks only, TLS 1.2+.
All code analysis is AST-based on your machine. No source code is ever transmitted.
Run scripts/airgap_proof.py; it installs a socket-level firewall and fails the test on any unexpected connection.
Every suggested package is checked against the real registry before install, so a hallucinated name never becomes a supply-chain hole.
No delete, deploy, or drop action ships without a human explicitly confirming first.
Compatible clients
nablr uses stdio transport, works with any MCP-compatible client.
Gemini and GitHub Copilot do not currently support MCP prompts.
Not a Claude Skill
Claude Code runs Skills and MCP servers side by side. They solve different problems - here's the actual difference.
Not a replacement, not a competitor. Claude Skills are Claude's own capability library. nablr is the governance layer sitting alongside it - run both in the same Claude Code session, zero conflict.
Read the full comparison →
