Run Pipeline

Run Pipeline

Execute and monitor Harness pipeline runs via MCP.

Instructions

Step 1: Find the Pipeline

Call MCP tool: harness_search
Parameters:
  query: ""
  resource_types: ["pipeline"]
  compact: true

Or list all pipelines in a project:

Call MCP tool: harness_list
Parameters:
  resource_type: "pipeline"
  org_id: ""
  project_id: ""
  search_term: ""

Step 2: Get Pipeline Details

Call MCP tool: harness_get
Parameters:
  resource_type: "pipeline"
  resource_id: ""
  org_id: ""
  project_id: ""

Extract from response:

• Required runtime inputs (`` fields)
• Pipeline variables with types and defaults
• Stage structure and deployment targets

Step 3: Check Available Input Sets

Call MCP tool: harness_list
Parameters:
  resource_type: "input_set"
  org_id: ""
  project_id: ""

Input sets provide pre-configured values that can be used or overridden.

Step 4: Execute the Pipeline

Call MCP tool: harness_execute
Parameters:
  resource_type: "pipeline"
  action: "run"
  resource_id: ""
  org_id: ""
  project_id: ""

Step 5: Monitor Execution

Call MCP tool: harness_get
Parameters:
  resource_type: "execution"
  resource_id: ""
  org_id: ""
  project_id: ""

Execution statuses: Running, Success, Failed, Aborted, Waiting (approval/input), Expired

Step 6: Handle Approvals

If execution is waiting for approval:

Call MCP tool: harness_list
Parameters:
  resource_type: "approval_instance"
  org_id: ""
  project_id: ""

Then approve or reject:

Call MCP tool: harness_execute
Parameters:
  resource_type: "approval_instance"
  action: "approve"  # or "reject"
  resource_id: ""

Safety: Production Deployments

CRITICAL: Before executing pipelines targeting production:

1. Confirm intent with the user explicitly
2. Show what will be deployed (version, service, environment)
3. Check if pipeline has approval gates
4. Verify the version was tested in lower environments

Response Format

Before Execution

Pipeline: 
Project: 
Required Inputs: 
Available Input Sets: 
Ready to execute? Confirm to proceed.

After Trigger

Pipeline: 
Execution ID: 
Status: Running
View in Harness: 

Execution Complete

Pipeline: 
Status: Success/Failed
Duration: 
Stages: 

Retrying Failed Executions

Retry is an action on pipeline (not execution). Pass the pipeline ID as resource_id and the failed execution's ID as execution_id:

Call MCP tool: harness_execute
Parameters:
  resource_type: "pipeline"
  action: "retry"
  resource_id: ""
  execution_id: ""
  org_id: ""
  project_id: ""

Interrupting Running Executions

To abort an entire pipeline execution:

Call MCP tool: harness_execute
Parameters:
  resource_type: "execution"
  action: "interrupt"
  resource_id: ""
  org_id: ""
  project_id: ""
  params:
    interrupt_type: "AbortAll"

The interrupt_type field inside params is required. Valid values:

| interrupt_type | Effect | |----------------|--------| | AbortAll | Abort the entire pipeline execution | | UserMarkedFailure | Mark the execution as failed by user |

To abort multiple stuck executions, call harness_execute with params: {interrupt_type: "AbortAll"} for each execution ID.

Examples

• "Run the ci-pipeline" - Search for pipeline, execute with defaults
• "Deploy version 2.0.0 to staging" - Find deploy pipeline, provide version input, execute
• "What's the status of execution xyz123?" - Get execution details
• "Retry the last failed deployment" - List recent failed executions, retry
• "Abort all stuck executions" - List running/waiting executions, interrupt each with params: {interrupt_type: "AbortAll"}
• "Stop execution abc123" - Interrupt the specific execution with params: {interrupt_type: "AbortAll"}

Performance Notes

• Always confirm the pipeline identifier and required inputs before triggering execution. A failed execution due to missing inputs wastes time.
• When monitoring executions, wait for terminal status (Success, Failed, Aborted) before reporting results. Do not assume intermediate states are final.
• If an execution fails, gather full error context from logs before suggesting fixes. Do not guess at root causes.

Troubleshooting

Pipeline Won't Start

• Check delegate availability with harness_status
• Validate all required inputs have values
• Test connector credentials with harness_execute (resourcetype: "connector", action: "testconnection")

Missing Inputs

• Use harness_get on the pipeline to see all `` fields
• Check input sets for pre-configured values

Execution Stuck

• Check for pending approvals with harness_list (resourcetype: "approvalinstance")
• Check delegate status with harness_status
• Abort stuck executions with harness_execute (resourcetype: "execution", action: "interrupt", params: {interrupttype: "AbortAll"})

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: harness
• Source: harness/harness-skills
• License: Apache-2.0
• Homepage: https://developer.harness.io/docs/platform/harness-ai/harness-skills/

Install and usage instructions live in the source repository linked above.