Harness MCP Server: run and debug Harness pipelines from your AI assistant

• Lists pipelines, services, flags, projects and other Harness items

  Lists any kind of Harness item, such as pipelines, executions, services, environments, feature flags or projects, with filters, search and paging.

  Try: “List the pipelines in the payments project”

  Network · harness_list

• Shows the details of one pipeline, run, service or other item

  Fetches one Harness item by its ID or a pasted Harness link, for example a pipeline's YAML or a single execution.

  Try: “Show me the YAML for the deploy-api pipeline”

  Network · harness_get

• Searches across all kinds of Harness items with one query

  Searches many Harness resource types at once with one query and returns matching items, so you do not need to know where something lives.

  Try: “Find anything related to the checkout service”

  Network · harness_search

• Explains why a pipeline run, connector or delegate is failing

  Looks into a pipeline run, connector, delegate or GitOps app and returns step timings, failure details and health signals so you can see what went wrong.

  Try: “Why did the last run of the payments pipeline fail?”

  Network · harness_diagnose

• Gives a health summary of a project's recent runs and failure rates

  Returns a quick health view of a project: recent executions, failure rates and links into the Harness UI.

  Try: “How healthy is the checkout project this week?”

  Network · harness_status

• Runs or retries pipelines, switches feature flags and syncs GitOps apps

  Carries out actions such as running or retrying a pipeline, importing a pipeline from Git, turning a feature flag on or off, or syncing a GitOps app. It asks you to confirm risky actions where your app supports it.

  Try: “Rerun the failed staging deployment”

  Runs code · harness_execute

• Creates pipelines, services, environments and other Harness items

  Creates a new Harness item, such as a pipeline, trigger, service or environment, including pipelines stored in Git.

  Try: “Create a pipeline that builds and deploys the api service to staging”

  Write · harness_create

• Changes an existing pipeline, service or other Harness item

  Updates an existing Harness item, for example editing a pipeline's stages or a service's settings.

  Try: “Add a manual approval step before production in the deploy-api pipeline”

  Write · harness_update

• Deletes a Harness item after you confirm

  Deletes a Harness item such as a pipeline or trigger. It asks for confirmation first and is blocked in apps that cannot show the prompt.

  Try: “Delete the old fx-rates pipeline”

  Delete · harness_delete

• Used by the assistant behind the scenes

  Lists what kinds of Harness items and actions are available

  Returns the resource types, operations and fields the server knows about, so the assistant can find the right one. It makes no call to Harness.

  Read · harness_describe

• Looks up the exact format for creating or editing an item

  Fetches the YAML or JSON schema and examples for a resource type, such as a pipeline or connector, so the assistant writes valid definitions.

  Network · harness_schema

Watch out for

• It can run pipelines and deploy

  It can run and retry pipelines, switch feature flags and sync GitOps apps. Use read-only mode or a key limited to the projects you need. github.com

• Some changes go through without asking

  Low-risk creates and updates run without a prompt, and Claude Desktop cannot show prompts yet. Leave HARNESS_AUTO_APPROVE_RISK at its default of none. github.com

• A path handling fix is still in progress

  A path traversal report from September led to a hardening change opened in late October. Keep the server on the newest version. github.com

• Secret values are never returned

  Secrets come back as names and metadata only, so the assistant cannot read their values. developer.harness.io

• The hosted server acts as you

  After you sign in it uses your Harness user and role permissions. developer.harness.io

Security advisories

In their words