Camunda C8ctl

Camunda c8ctl CLI

Install and use c8ctl — the minimal-dependency CLI for Camunda 8.8+ — for connecting to clusters, deploying resources, and managing process automation. c8ctl is the foundation for the other camunda-* skills (camunda-bpmn, camunda-connectors, camunda-feel, camunda-process-mgmt, camunda-ai-agents).

Prerequisites

• Node.js ≥ 22.18.0 (required for native TypeScript support) — see camunda-development for installing it locally

Cross-References

• camunda-bpmn: Uses c8ctl bpmn lint
• camunda-connectors: Uses c8ctl element-template search/info/get-properties/apply
• camunda-feel: Uses c8ctl feel evaluate
• camunda-process-mgmt: Uses c8ctl deploy, c8ctl run, c8ctl watch, c8ctl list pi, c8ctl search inc, c8ctl complete ut, c8ctl resolve inc, etc.
• camunda-ai-agents: Uses c8ctl element-template search/apply to apply the AI Agent connector template

Instructions

Quick readiness check (onboarding)

Use this quick check when a user is new to c8ctl and you want to confirm the machine is ready before deeper setup:

# Required runtime
node --version 2>/dev/null || echo "NOT FOUND"

# Required CLI
c8ctl --version 2>/dev/null || echo "NOT FOUND"

# Required default plugins for camunda-* skills
c8ctl bpmn --help >/dev/null 2>&1 && echo "bpmn: OK" || echo "bpmn: MISSING"
c8ctl element-template --help >/dev/null 2>&1 && echo "element-template: OK" || echo "element-template: MISSING"
c8ctl feel --help >/dev/null 2>&1 && echo "feel: OK" || echo "feel: MISSING"

# Optional: check current active profile
c8ctl which profile 2>/dev/null || echo "no active profile"

# Optional: check whether a local c8run cluster is already running
c8ctl cluster status 2>/dev/null || echo "local cluster: not running"

• Node.js and c8ctl are required. Node must be ≥ 22.18.0. If either is missing (or Node is older), install/upgrade before continuing.
• If any default plugin is missing, the installed c8ctl is older than 3.0.0. Ask the user to confirm before upgrading — don't run the install unprompted — then run npm install -g @camunda8/cli.
• Missing profile/local cluster is not an error; it only means profile or cluster setup is still pending.

Install

Install c8ctl globally from npm. The other camunda-* skills depend on the bpmn, element-template, and feel plugins, which require c8ctl ≥ 3.0.0:

npm install -g @camunda8/cli

After installation, both c8ctl and the shorter alias c8 are available. The other camunda-* skills use the c8ctl form for clarity.

Verify:

c8ctl --version
c8ctl help

Verify default plugins

The other camunda-* skills depend on three plugins that ship with c8ctl ≥ 3.0.0: bpmn, element-template, and feel. Verify each is available:

c8ctl bpmn --help              # camunda-bpmn
c8ctl element-template --help  # camunda-connectors, camunda-ai-agents
c8ctl feel --help              # camunda-feel

If any command exits non-zero, the installed c8ctl is older than 3.0.0 and lacks these plugins. Ask the user to confirm before installing — don't run the install unprompted — then run:

npm install -g @camunda8/cli

Pick a Cluster

c8ctl needs a cluster to talk to. Before configuring a profile, ask the user which cluster they want to use:

1. Local development cluster (c8run) — recommended default for new projects, experiments, and local iteration. c8ctl can download, start, and manage c8run for you. See "Local Cluster" below.
2. Camunda 8 SaaS — managed cluster in Camunda's cloud. The user provides client credentials.
3. Self-Managed — a Camunda 8 cluster the user runs themselves (Kubernetes, Docker Compose, etc.). The user provides the base URL and (if secured) OAuth credentials.
4. Camunda Modeler profile — if the user has Camunda Desktop Modeler installed with a configured connection, c8ctl auto-imports those profiles. Use them with the modeler: prefix (e.g., modeler:Local Dev).

If the user hasn't decided and is doing local development, suggest local c8run — it has zero setup cost beyond c8ctl cluster start.

Local Cluster (c8ctl cluster)

c8ctl ships with a default cluster plugin that wraps c8run. It downloads, starts, and stops a local Camunda 8 cluster for you.

Before starting a new local cluster, run a quick preflight so you don't restart unnecessarily and you catch Java/runtime issues early:

# Is a local cluster already responding?
curl -s -o /dev/null -w "%{http_code}" http://localhost:8080/v2/topology

# Is Java available (c8run needs JRE 21+)?
java -version

• If topology returns 200, a local cluster is already running; continue with your task.
• If Java is missing or too old, install JRE/JDK 21+ first.
• In some non-interactive shells, java -version can work while JAVA_HOME is still unset (common with asdf/mise shims). If startup fails even though Java is installed, set JAVA_HOME explicitly before c8ctl cluster start:

export JAVA_HOME="$(asdf where java 2>/dev/null || mise where java 2>/dev/null)"

Examples:

# Start the latest stable release (downloads on first run)
c8ctl cluster start

# Start a specific minor version (rolling release — auto-updates patches)
c8ctl cluster start 8.9

# Start a specific full version
c8ctl cluster start 8.9.0-alpha5

# Use the alpha alias for the latest pre-release
c8ctl cluster start alpha

# Check status (running? what version? connection details?)
c8ctl cluster status

# Verify the cluster is responding after start
c8ctl get topology

# Stream logs
c8ctl cluster logs

# Stop the cluster
c8ctl cluster stop

The cluster runs on http://localhost:8080 by default. With no profile configured, c8ctl falls back to http://localhost:8080/v2, so most commands work out of the box against a freshly started local cluster.

For full c8ctl cluster command reference (list, install, list-remote, delete, cache locations, version aliases), see references/local-cluster.md.

Connect: Profiles

Profiles store cluster connection details. A local profile shipped with c8ctl already points at http://localhost:8080/v2, so no setup is needed for local c8run work — just pass --profile=local. Camunda Desktop Modeler profiles are auto-imported with a modeler: prefix.

For OAuth-secured clusters (SaaS or Self-Managed), profile setup is a one-time human task — see the c8ctl docs for add profile flags and credential handling. Don't run add profile on the agent's initiative; ask the user to configure profiles before using this skill.

Use already-configured profiles:

# Switch the active profile
c8ctl use profile prod

# Show the current active profile
c8ctl which profile

# List all profiles (includes Camunda Modeler profiles with `modeler:` prefix)
c8ctl list profiles

# Use a Modeler profile
c8ctl use profile "modeler:Local Dev"

# One-off override for a single command (active profile unchanged)
c8ctl list pi --profile=staging

For multi-tenancy, credential resolution order, and Modeler integration details, see references/profiles.md.

Safety: target the right cluster

c8ctl resolves cluster connections via the globally-active profile by default. That's risky: the active profile might still point at production or staging from a previous session, and a cluster-touching command will silently target it.

Always pass --profile=<name> explicitly on commands that touch a cluster, especially mutating ones (deploy, run, cancel, resolve, complete, publish, watch). Read-only commands (get, list, search, feel evaluate) are safer but the same discipline keeps the transcript auditable and prevents a forgotten c8ctl use profile prod from silently steering the next command.

Session opener: before doing any cluster-touching work, run c8ctl which profile and confirm with the user. If the active profile name suggests a shared environment (prod, production, staging, live, saas-prefixed, customer names), ask before acting — don't assume that profile is intended for the current work.

For new local-development projects, use --profile=local — don't inherit whatever profile a previous project left active.

c8ctl cluster start/stop/status/logs are unaffected: they operate on the local c8run process directly, not via a profile.

Use the CLI

c8ctl has two command shapes:

• Core API commands follow <verb> <resource> — list pi, get inc <key>, complete ut <key>. Resources have short aliases (pi = process-instance, pd = process-definition, ut = user-task, inc = incident, msg = message).
• Plugin commands follow <plugin> <subcommand> — cluster start, element-template apply, bpmn lint, feel evaluate. The plugin name is the first token; subcommands are plugin-defined.

Quick tour (examples omit --profile=<name> for brevity — pass it explicitly per the Safety rule above):

# Inspect the cluster
c8ctl get topology
c8ctl list pd                                  # deployed process definitions
c8ctl list pi                                  # running process instances
c8ctl search inc --state=ACTIVE                # active incidents

# Develop / deploy
c8ctl deploy ./process.bpmn                    # deploy a single resource
c8ctl deploy ./my-project                      # deploy a directory (supports _bb-* and .process-application)
c8ctl run ./order.bpmn --variables='{"orderId":"42"}'   # deploy + start in one step
c8ctl watch                                    # auto-redeploy on file save
c8ctl await pi --id=order-process              # block until completion

# Operate
c8ctl get pi 2251799813685249 --variables      # inspect an instance
c8ctl complete ut 2251799813685250 --variables='{"approved":true}'
c8ctl publish msg payment-received --correlationKey=order-42
c8ctl resolve inc 2251799813685251
c8ctl cancel pi 2251799813685249

For the full verb/resource matrix, plugin commands, and resource-specific flags, see references/command-reference.md.

Output Modes (for AI / scripting)

For AI-driven and scripted use, request JSON per command and combine with --fields for stable structured output:

# Per-command JSON (deterministic, does not mutate session state)
c8ctl list pd --json --fields=key,bpmnProcessId,version,name

# Preview an API request without executing it
c8ctl deploy ./process.bpmn --dry-run

# Pagination / sort (on list and search commands)
c8ctl list pi --limit=50 --sortBy=startDate --desc

Prefer the per-invocation --json flag over c8ctl output json — the latter mutates session.json and leaks across other tools and sessions. The C8CTL_OUTPUT_MODE=json env var works for the current shell too, but shell state does not persist across separate tool calls in most agent harnesses, so an export in one step won't carry to the next. Always pass --json on the command itself when you need structured output.

Plugins

Default plugins (cluster, bpmn, feel, element-template) ship pre-loaded with c8ctl — no install step needed. List them with c8ctl list plugins. For installing or managing additional plugins, see the c8ctl docs.

Don't run c8ctl load plugin on the agent's initiative — plugin code runs in the c8ctl runtime with full access to the active profile. Ask the user to install plugins themselves.

Troubleshooting

• c8ctl: command not found (or c8: command not found) — npm's global bin directory isn't on PATH. Run npm config get prefix and add <prefix>/bin to PATH.
• Node.js version too old — c8ctl requires Node ≥ 22.18.0 for native TypeScript support. Use nvm or asdf to upgrade.
• Local cluster won't start — check c8ctl cluster status and c8ctl cluster logs. Common causes: port 8080 already in use, Java not installed (c8run needs JRE 21+), insufficient disk space for the binary download.
• c8ctl cluster start reports "port 8080 in use" but the port is actually free (lsof/nc show nothing listening) — sandboxed environments that block socket binding (some coding-agent harnesses, restricted container modes, macOS App Sandbox) surface this way. Ask the user to run c8ctl cluster start directly in their host terminal (outside the agent sandbox).
• c8ctl can't write to its default data directory (sandboxed agents, restricted filesystems) — set C8CTL_DATA_DIR=<writable-path> before invoking c8ctl.
• OAuth errors against SaaS — verify the profile is configured correctly. The cluster URL for SaaS is the Zeebe REST address, not the dashboard URL. See the c8ctl docs for OAuth flags.

References

For detailed reference material, read from references/:

• local-cluster.md — full c8ctl cluster command reference, version aliases (stable/alpha/rolling), cache locations, connector-secrets bootstrap flow
• profiles.md — profile management, OAuth flags, Modeler integration, tenant resolution, credential resolution order, environment variables
• command-reference.md — verb/resource matrix, plugin command shape, resource aliases, search flags, global flags, default plugins