> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nablrco.com/llms.txt
> Use this file to discover all available pages before exploring further.

# nablr vs. Claude Skills

> nablr is not a Claude Skill - here's the actual difference

<Card title="Short answer" icon="circle-check">
  Claude Skills are prompt+script bundles Claude loads into its own context and decides to trigger. nablr is a separate MCP server - a stateful SDLC orchestration layer any MCP client can talk to, not just Claude.
</Card>

## <Icon icon="triangle-alert" /> Why people ask

Claude Code supports both **Skills** and **MCP servers** side by side, and both show up as "things Claude can use." That's where the confusion starts. They solve different problems.

## <Icon icon="git-compare" /> Side by side

|                              | Claude Skills                                                                 | nablr                                                                                                                                                     |
| ---------------------------- | ----------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **What it is**               | A `SKILL.md` + scripts bundle loaded into Claude's own context                | An external MCP server process Claude (or any MCP client) connects to over stdio                                                                          |
| **Who triggers it**          | Claude, by matching the task against the skill's description                  | You, explicitly (`Set agent to developer`) - or nablr's own workflow state                                                                                |
| **Works with**               | Claude only (Claude.ai, Claude Code, Claude Desktop)                          | Cursor, Claude Desktop, Windsurf - any MCP-compatible client                                                                                              |
| **State across turns**       | None - each invocation is a stateless prompt injection                        | Persistent session/story state (`active_agent`, `active_story`, blockers, halt flags) tracked across the whole workflow                                   |
| **Enforcement**              | Advisory - Claude follows the skill's instructions if it decides to invoke it | Gated - `qa_lead` validates acceptance criteria before `developer` can start, `pr_reviewer` checks compliance before merge, blockers escalate at L1/L2/L3 |
| **Analysis**                 | Whatever the skill's scripts happen to do                                     | 40+ local AST scanners (dead code, complexity, security, tech debt) shipped as first-class tools                                                          |
| **Multi-step orchestration** | Skills don't sequence each other                                              | 19 personas in a defined handoff sequence: product\_manager → qa\_lead → architect → developer → tester → pr\_reviewer                                    |
| **Licensing model**          | N/A                                                                           | Free tier (scanners, workflow, codemap) + Pro tier (personas, rule files) via license key                                                                 |
| **Audit trail**              | None built in                                                                 | Structured event logs for governance / compliance evidence                                                                                                |

## <Icon icon="layers" /> They're not competitors

nablr doesn't replace Claude Skills, and it isn't trying to be one. A Claude Skill is a capability Claude reaches for inside its own context. nablr is the governance layer sitting alongside your assistant, enforcing quality gates and holding workflow state that no skill - stateless by design - can hold on its own.

You can run nablr and Claude Skills in the same Claude Code session with no conflict.

## <Icon icon="cog" /> How to tell which one you're looking at

* Lives in `.claude/skills/` or a plugin, described in a `SKILL.md` → **Claude Skill**
* Shows up in your MCP client config (`mcp.json`, `claude_desktop_config.json`) as a `command` + `args` → **MCP server** (this is nablr)

<Note>Still unsure whether nablr fits your workflow? See [When to use nablr](/introduction#-when-to-use-nablr) or try a [use case](/use-cases/code-to-clarity).</Note>