commerce-tools

A modular AI tooling platform for Adobe Commerce operations, monitoring, and security analysis. Provides New Relic and Jira MCP servers, plus a composable framework of agents, skills, commands, and providers for building AI-powered Commerce workflows.

Features

• New Relic MCP Server — executes NRQL queries via NerdGraph and maps Adobe Commerce project IDs to New Relic account IDs
• Jira MCP Server — fetches a Jira issue by key and extracts any New Relic account/project id mentioned in its description or comments
• Jira → New Relic Triage Workflow — a work-on-ticket command and jira-newrelic-triage skill that chain the two MCP servers to investigate a ticket end-to-end
• Commerce & Security Prompts — built-in specialist prompts for APM monitoring, performance analysis, and attack detection (brute force, SQLi, XSS, DDoS, scraping, carding)
• Agent Framework — automatic tool-use loop with conversation history management and configurable iteration limits
• Pluggable Providers — abstract provider interface supports Claude and other LLM backends
• Hook System — event-driven extensibility with before-tool, after-tool, before-generate, and after-generate hooks
• Composable CLI — command, skill, and agent registries for building and extending tools

Project Structure

commerce-tools/
├── agents/        # Agent implementations (extend BaseAgent)
├── commands/      # CLI command registry
├── core/          # Runtime: router, tool-runner, session, hooks
├── mcps/          # MCP servers (New Relic NRQL, Jira)
├── plugins/       # Claude/Cursor plugin build (commands, skills, mcp config)
├── providers/     # AI provider implementations and base types
├── skills/        # Skill registry and loaders
├── dist/          # Compiled output (generated)
├── .env.example   # Environment variable template
└── .mcp.json      # MCP server config for Claude Code

Getting Started

Prerequisites

• Node.js 18+
• A New Relic account with a NerdGraph API key
• A Jira Cloud site with an API token (only needed for the Jira MCP server)

Installation

npm install

Configuration

Copy .env.example to .env and fill in your credentials:

cp .env.example .env

| Variable | Description | |---|---| | NEW_RELIC_API_KEY | NerdGraph API key | | NEW_RELIC_REGION | us or eu (default: us) | | NEW_RELIC_ACCOUNT_ID | Optional default account ID | | JIRA_BASE_URL | Jira site URL, e.g. https://yourcompany.atlassian.net | | JIRA_EMAIL | Email of the account owning the API token | | JIRA_API_TOKEN | Jira API token | | MCP_TRANSPORT | Transport mode (default: stdio) |

Build

npm run build

Run the New Relic MCP server

npm start
# or directly:
node dist/mcps/newrelic/server.js

Run the Jira MCP server

npm run start:jira
# or directly:
node dist/mcps/jira/server.js

Using with Claude Code

The .mcp.json file configures both the New Relic and Jira MCP servers for use with Claude Code. To register them manually instead:

claude mcp add newrelic -- node /path/to/commerce-tools/dist/mcps/newrelic/server.js
claude mcp add jira -- node /path/to/commerce-tools/dist/mcps/jira/server.js

Or reference the existing .mcp.json in your Claude Code project settings. Once connected, Claude can:

• Run NRQL queries against any New Relic account
• Resolve Adobe Commerce project IDs to account IDs
• Analyze logs, transactions, APM metrics, Redis, MySQL, and Elasticsearch data
• Detect and investigate security attacks
• Fetch a Jira issue by key and pull its summary, status, description, and comments
• Chain a Jira ticket straight into a New Relic investigation via the jira-newrelic-triage workflow below

Architecture

Core

| Module | Purpose | |---|---| | core/router.ts | Resolves provider names to implementations | | core/tool-runner.ts | Registers and dispatches tool handlers | | core/session.ts | Manages conversation sessions and message history | | core/hooks.ts | Event hooks for tool and generation lifecycle |

Agents

agents/_base.ts implements the agentic loop:

1. Call provider.generate() with current messages and tools
2. Append assistant response to history
3. If stop reason is tool_use, extract and execute tool calls
4. Append tool results to history
5. Repeat up to maxIterations (default: 10)

Subclass BaseAgent and implement handleToolCall() to build custom agents.

New Relic MCP Tools

| Tool | Description | |---|---| | execute_nrql | Run a NRQL query against a New Relic account via NerdGraph | | get_account_id_by_project_id | Resolve an Adobe Commerce project ID to a New Relic account ID |

New Relic MCP Prompts

| Prompt | Description | |---|---| | newrelic_commerce | NRQL specialist for Adobe Commerce monitoring and investigation | | newrelic_security | Extends commerce prompt with attack detection and security analysis |

Jira MCP Tools

| Tool | Description | |---|---| | get_jira_issue | Fetch a Jira issue by key (e.g. PRO-234): summary, status, description, comments, and a best-effort extraction of a New Relic account/project id mentioned in the text |

Jira → New Relic Triage Workflow

The Jira MCP is paired with a prompt (mcps/jira/prompts/workflow.ts) that drives an end-to-end investigation of a ticket:

1. Fetch the ticket via get_jira_issue.
2. Check extractedIds in the result — if an accountId or projectId was found in the description/comments, use it. If both are empty, the workflow asks the user for one rather than guessing.
3. Resolve the New Relic account, calling get_account_id_by_project_id first if only a project ID is known.
4. Investigate with NRQL via execute_nrql, following the newrelic_commerce prompt's field-discovery rules.
5. Summarize findings back to the user, referencing the ticket key/summary. The ticket itself is never modified — no comments or status changes are made.

This workflow ships as a Claude Code plugin (see plugins/claude/):

• Command — /work-on-ticket <issue-key> runs the workflow directly for a given ticket.
• Skill — jira-newrelic-triage is auto-invoked when you mention working on, investigating, or triaging a Jira ticket (e.g. "work on jira ticket PRO-234").

Both are generated from mcps/jira/prompts/workflow.ts by plugins/build.ts (npm run build:plugins) — edit the source prompt, not the generated files under plugins/*/commands and plugins/*/skills.

Development

# Watch mode
npx tsc --watch

# Type check only
npx tsc --noEmit

License

MIT