Artillery MCP

@kosiakmd/artillery-mcp

Richer drop-in MCP server for Artillery 2.x — exposes the full artillery run and artillery run-fargate flag surface, parses raw text output (artillery-output.txt), returns every counter/rate/summary Artillery produces (not just http.*), and optionally adds a config-driven project launcher and counter-grouping for your custom counters.

Why another MCP server?

The upstream @jch1887/artillery-mcp-server was a solid start but missed several flags our team needed day-to-day. This package is a MIT-licensed fork with broader coverage. See Credits.

Feature	upstream	this package
run_test_from_file flags	path + output only	+ --record/--key/--tags/--name/--note/-t/-e/--scenario-name/-v/--overrides/-p/--dotenv/-k/--count/-s
AWS Fargate (run-fargate)	❌	✅ full flag set
parse_results counters	http.* only	ALL counters + rates + nested summaries
Parse raw stdout (artillery-output.txt)	❌	✅ read_artillery_output tool
HTML report generation	passes invalid --report flag	✅ separate artillery report call
Standalone JSON → HTML (run_report)	❌	✅
--dry-run validation	broken (flag doesn't exist in 2.x)	✅ client-side YAML structural check
Opt-in project launcher	❌	✅ run_project_lt (when config present)
Opt-in counter grouping	❌	✅ counterBreakdown (when config present)
init scaffolder	❌	✅ npx @kosiakmd/artillery-mcp init
Shipped SKILL.md template	❌	✅ agent-oriented guidance
Docker image	❌	✅ multi-arch (Docker Hub + GHCR)
MCP Registry listing	❌	✅ io.github.kosiakMD/artillery-mcp
serverVersion accuracy	hardcoded	✅ read from package.json at runtime
Tests	122	163

Quickstart for agents (Claude Code / Cursor)

# In your project root
npx @kosiakmd/artillery-mcp init

Scaffolds two files (skipped if they already exist, use --force to overwrite):

• .artillery-mcp.config.json — starter template with commented fields
• .ai/skills/artillery-mcp/SKILL.md — agent-oriented guidance (when to call which tool, common patterns, gotchas)

After editing the config with your real flows/paths, register the MCP in your agent and restart the session:

claude mcp add artillery-mcp -s user \
  -e ARTILLERY_WORKDIR="$PWD" \
  -e ARTILLERY_CLOUD_API_KEY=a9_... \
  -- npx -y @kosiakmd/artillery-mcp

Install / Quickstart (zero-config)

Runs out of the box — no config file needed. Works as an MCP server over stdio for Claude Code, Claude Desktop, Cursor, and any MCP-compatible client.

# Using npx (no install)
npx -y @kosiakmd/artillery-mcp

# Or install globally
npm install -g @kosiakmd/artillery-mcp
artillery-mcp

Claude Code / Cursor — register as an MCP server in your client config:

{
  "mcpServers": {
    "artillery-mcp": {
      "command": "npx",
      "args": ["-y", "@kosiakmd/artillery-mcp"],
      "env": {
        "ARTILLERY_WORKDIR": "/absolute/path/to/your/project",
        "ARTILLERY_CLOUD_API_KEY": "a9_..."
      }
    }
  }
}

Claude Code CLI:

claude mcp add artillery-mcp -s user \
  -e ARTILLERY_WORKDIR=/abs/path/to/project \
  -e ARTILLERY_CLOUD_API_KEY=a9_... \
  -- npx -y @kosiakmd/artillery-mcp

Prerequisites: Node.js ≥ 20; Artillery CLI on PATH (npm i -g artillery).

Docker

Multi-arch image — linux/amd64 and linux/arm64, Artillery CLI preinstalled. ~500 MB (Chromium/Playwright browsers skipped — see below). Published to both Docker Hub and GitHub Container Registry from the same build; identical digests.

# Docker Hub (discoverable via `docker search artillery-mcp`)
docker pull kosiakmd/artillery-mcp:latest

# GitHub Container Registry
docker pull ghcr.io/kosiakmd/artillery-mcp:latest

Run (mount your project as /workspace):

docker run -i --init --rm \
  -v "$PWD":/workspace \
  -e ARTILLERY_CLOUD_API_KEY="$ARTILLERY_CLOUD_API_KEY" \
  kosiakmd/artillery-mcp:latest

--init ensures the Node process gets reaped on stdin close. Mount /workspace read-write if you want save_config tools to persist to /workspace/saved-configs/; read-only is fine otherwise (the save-config family will simply return errors when called).

MCP client config (Claude Desktop / Cursor) — point the MCP at docker instead of npx:

{
  "mcpServers": {
    "artillery-mcp": {
      "command": "docker",
      "args": [
        "run", "-i", "--init", "--rm",
        "-v", "/absolute/path/to/your/project:/workspace",
        "-e", "ARTILLERY_CLOUD_API_KEY",
        "kosiakmd/artillery-mcp:latest"
      ],
      "env": { "ARTILLERY_CLOUD_API_KEY": "a9_..." }
    }
  }
}

Playwright engine? If you use engine: playwright in your Artillery scripts, extend the base image with Chromium:

FROM kosiakmd/artillery-mcp:latest
ENV PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=0
RUN apk add --no-cache chromium nss freetype harfbuzz ttf-freefont \
 && npm i -g @playwright/test \
 && npx playwright install chromium

Image tags:

• latest — most recent release
• v0.1.1, v0.1, v0 — pinned by semver (patch / minor / major)

Environment variables

Var	Purpose	Default
ARTILLERY_BIN	Path to artillery binary	auto-detected via which artillery
ARTILLERY_WORKDIR	Working directory for runs	cwd
ARTILLERY_TIMEOUT_MS	Max duration of a single run	1800000 (30 min)
ARTILLERY_MAX_OUTPUT_MB	Cap on captured stdout/stderr	10
ARTILLERY_ALLOW_QUICK	Enable quick_test tool	true
ARTILLERY_CLOUD_API_KEY	Used when a tool is called with record: true	—
ARTILLERY_MCP_CONFIG	Absolute path to project config (opt-in)	—
DEBUG	artillery:mcp:* for verbose logs	—

18 base tools (no config needed)

Run tests

• run_test_from_file — full-flag artillery run wrapper (see flag surface above).
• run_test_inline — same but takes YAML text, writes to a tmp file.
• quick_test — artillery quick <url> with rate/count/duration/method/headers/body.
• run_fargate — artillery run-fargate with --region, --cluster, --cpu, --memory, --launch-type, --spot, --subnet-ids, --security-group-ids, --task-role-name, --task-ephemeral-storage, --container-dns-servers, --max-duration, --packages, --secret, --no-assign-public-ip + all run-shared flags.
• run_preset_test — smoke / baseline / soak / spike presets against a URL.
• run_saved_config — re-run a saved config by name.

Parse + inspect results

• parse_results — reads an Artillery JSON report; returns summary + allCounters + allRates + allSummaries + scenarios + metadata.
• read_artillery_output — reads a raw Artillery stdout dump (e.g. artillery-output.txt saved from CI), returns rawText (with tail-truncation), summaryBlock, counters, rates, metrics (nested percentiles).
• run_report — converts an existing JSON results file to HTML via artillery report. Use when you have JSON from CI artifacts and want shareable HTML without re-running.
• compare_results — diff two Artillery JSON results for regression detection.

Saved configs

• save_config / list_configs / get_config / delete_config — persistent named Artillery configs.

Interactive builder

• wizard_start / wizard_step / wizard_finalize — interactive test builder.

Meta

• list_capabilities — versions of Artillery/Node/this server, configured paths, limits.

Optional feature #1 — Project launcher (run_project_lt)

Activated when a .artillery-mcp.config.json file is discovered. Gives you one-liner invocations instead of writing full artillery run ... flag lists.

Discovery precedence:

1. ARTILLERY_MCP_CONFIG env var (absolute path)
2. Walk up from ARTILLERY_WORKDIR / cwd looking for .artillery-mcp.config.json
3. No config → this tool is not registered

Config (.artillery-mcp.config.json):

{
  "flows": {
    "browse": "tests/load/browse.yml",
    "buy": "tests/load/buy.yml",
    "combined": "tests/load/combined.yml"
  },
  "environments": ["local", "staging", "prod"],
  "defaultTags": { "repo": "my-app", "owner": "Platform" },
  "tagTemplates": ["type:{flow}", "env:{env}", "source:mcp"],
  "outputDir": "load-test-results"
}

Call:

{ "flow": "buy", "environment": "staging", "note": "canary v42" }

Effective command:

artillery run \
  --record \
  --name "buy-staging-2026-04-15T00-35-00-000Z" \
  --tags "repo:my-app,owner:Platform,type:buy,env:staging,source:mcp" \
  -e staging \
  --note "canary v42" \
  -o /abs/project/load-test-results/buy-staging-2026-04-15T00-35-00-000Z.json \
  /abs/project/tests/load/buy.yml

Optional fields:

• tagTemplates — {flow}, {env}, plus any caller-supplied templateVars (e.g. {round})
• outputDir — relative to project root; if missing, outputs land at project root
• defaultTags — merged before templates
• environments — if empty, any environment name is accepted

Override on the call side: name, note, extraTags, outputJson, reportHtml, variables, overrides, record: false, validateOnly, extraArgs, templateVars.

Optional feature #2 — Counter-group bucketing (counterBreakdown)

When you emit custom counters via events.emit('counter', 'shop.step.add_item.happy', 1), Artillery aggregates them into aggregate.counters. This server returns ALL of them in allCounters. Add counterGroups to your config to also get a pre-bucketed counterBreakdown in parse_results and read_artillery_output responses — ideal for CI pass/fail views and semantic grouping.

Config:

{
  "counterGroups": {
    "name": "shopBreakdown",
    "prefix": "shop.",
    "buckets": [
      { "key": "steps",   "match": "^shop\\.step\\." },
      { "key": "cart",    "match": "^shop\\.cart\\." },
      { "key": "payment", "match": "^shop\\.payment\\." },
      { "key": "flow",    "match": "^shop\\.flow\\." },
      { "key": "other",   "default": true }
    ]
  }
}

Example response (parse_results):

{
  "summary": { "requestsTotal": 1500, "rpsAvg": 25, "latencyMs": { "p95": 850 }, "errors": {} },
  "allCounters": { "shop.step.add_item.happy": 90, "http.requests": 1500, /* ... */ },
  "allRates": { "http.request_rate": 25 },
  "allSummaries": { "http.response_time": { "min": 45, "p99": 1800 } },
  "counterBreakdown": {
    "steps": { "shop.step.add_item.happy": 90, "shop.step.add_item.fail": 2 },
    "cart": { "shop.cart.failure.quantity_update": 1 },
    "payment": {},
    "flow": { "shop.flow.started": 100 },
    "other": {}
  }
}

Rules:

• prefix (optional) — counters not starting with this string are ignored
• buckets — ordered; first match wins
• One bucket may have default: true — catches everything that didn't match
• Invalid regex → throws at parse time (fix your config)

Without counterGroups, the counterBreakdown field is simply absent from responses.

Full config reference

{
  "flows": { "<name>": "<relative yaml path>" },  // enables run_project_lt
  "environments": ["<name>", "..."],              // optional whitelist
  "defaultTags": { "<k>": "<v>" },
  "tagTemplates": ["type:{flow}", "env:{env}"],
  "outputDir": "load-test-results",
  "counterGroups": {                              // enables counterBreakdown
    "name": "<output field name, cosmetic>",
    "prefix": "<optional prefix filter>",
    "buckets": [
      { "key": "<name>", "match": "<regex>" },
      { "key": "<name>", "default": true }
    ]
  }
}

Security

• No network I/O — only spawns the local artillery CLI.
• No eval / Function / dynamic imports.
• No install hooks.
• Dependencies pinned via package-lock.json (committed).
• Published with npm --provenance (SLSA attestation).

Roadmap

Shipped

• v0.1.1 — Docker image (multi-arch amd64/arm64; Artillery CLI preinstalled)
• v0.1.2 — artillery-mcp init scaffolder + shipped SKILL.md agent-guidance template
• v0.1.3 — hard-fail on unknown CLI args, fix Dockerfile missing skills/
• v0.1.4 — dual-publish to Docker Hub alongside GHCR (for docker search discoverability)
• v0.1.7 — run_report tool (JSON → HTML via artillery report); listed in MCP Official Registry (io.github.kosiakMD/artillery-mcp)

v0.2 (next)

• Artillery Cloud API integration — list_recent_runs, get_run_details(runUrl), compare_to_baseline(runUrl). Requires reverse-engineering the artilleryio REST API or partnering with Artillery.io.
• Config schema validation on startup — parse .artillery-mcp.config.json through zod with human-readable error messages ("expected 'flows' to be object, got null at line 3"). Fail-fast with pointer to README.
• YAML config support — accept .artillery-mcp.config.yml using a tiny bundled YAML parser (keeping deps light).
• Playwright Docker variant — kosiakmd/artillery-mcp:latest-playwright with Chromium preinstalled for users with engine: playwright scripts. Separate tag to keep base image small.

v0.3+

• Artillery Lambda + Azure ACI tools — run_lambda, run_aci for parity with run_fargate.
• run_project_lt matrix mode — {"matrix": {"flow": ["free","paid"], "env": ["staging","prod"]}} → 4 runs in parallel.
• Per-flow config overrides — different counterGroups / defaultTags / environments per flow instead of global.
• Streaming intermediate metrics via MCP progress events so agents see RPS/errors mid-run instead of only at completion.
• Built-in presets library — smoke / baseline / soak / spike selectable via config (not just inline YAML).
• Published JSON Schema at stable URL for IDE autocompletion of .artillery-mcp.config.json.
• Plugin API for counter-group matchers beyond regex — e.g. JSONPath, Wasm filter, callback to a user-provided JS.
• Structured run_report variant — return extracted summary text alongside the HTML path, so AI agents can skip loading the file.
• Smithery support — add StreamableHTTP transport + hosted deployment once there's enough demand (requires OAuth-style session config for per-user API keys).

Under consideration

• Grafana / Prometheus integration — push metrics to a user-provided Prometheus endpoint instead of (or in addition to) Artillery Cloud.
• Native Artillery Pro support — enterprise features if users request them.
• TUI dashboard for long-running tests when invoked outside MCP (standalone mode).
• Community MCP catalogs — manual registration on mcp-get.com, PulseMCP for extra discoverability.
• Upstream contribution — offer read_artillery_output, run_fargate, and full flag surface patches back to @jch1887/artillery-mcp-server as PRs; if merged, this fork becomes a thin config-plugin layer on top.

Issues / feature requests / PRs welcome: github.com/kosiakMD/artillery-mcp/issues.

Issues and feature requests welcome: github.com/kosiakMD/artillery-mcp/issues.

Credits

Forked from jch1887/artillery-mcp-server (MIT). See NOTICE for the list of additions and modifications.

License

MIT — see LICENSE. Both the upstream and this fork's copyright notices must be preserved in substantial portions of the software.