MCP SERVER · OFFICIAL PROJECTDrive a Kestra instance from your assistant — run flows, chase executions, replay failures, manage namespace files and KV pairs.
LABSCO SUMMARY
Most orchestration integrations stop at listing runs. Here the recovery path is covered too — replay an execution, restart from the failed task, resume something that is waiting, or backfill a period that never ran — which is the work you do at the moment something breaks, and exactly when you do not want to be clicking through a UI. Two configuration notes save time: on OSS, disable the EE group or the model will offer tools your instance cannot serve, and the container needs a host address it can actually reach, which on macOS and Windows means host.docker.internal rather than localhost.
Kestra's own Python MCP server, published as a container image, exposing the orchestration platform's API to an MCP client.
A reachable Kestra API in KESTRA_BASE_URL and a tenant in KESTRA_TENANT_ID. Authentication depends on edition: Enterprise uses KESTRA_API_TOKEN, while OSS installations can use KESTRA_USERNAME and KESTRA_PASSWORD. Runs from the published container image, so no Python environment to manage. OSS users should set KESTRA_MCP_DISABLED_TOOLS to ee, since those tools have nothing to talk to.
One command plus credentials — docker run -i --rm --pull always -e KESTRA_BASE_URL -e KESTRA_TENANT_ID -e KESTRA_MCP_DISABLED_TOOLS -e KESTRA_MCP_LOG_LEVEL -e KESTRA_USERNAME -e KESTRA_PASSWORD ghcr.io/kestra-io/mcp-server-python:latest, then supply credentials
USE CASES
Pull the execution and its logs and read the failing task's output directly in the conversation.
Restart from the task that failed, or replay the whole execution, once you have fixed the cause.
Trigger the backfill over the date range rather than starting each execution by hand.
Read and set the key-value entry that the flow picks up, without leaving the assistant.
Disable the Enterprise group so the model is not offered tools your edition cannot run.
Listed in project code.
backfill_executionsBackfill executions for a flow within a time window. You can provide explicit start/end dates or use the days/hours/minutes parameters to calculate the start/end dates relative to the current time. Provide those values as integers, e.g., 2 to backfill executions for the last 2 days/hours/minutes.
invite_userInvite a user to a tenant and and assign either a group or an IAM role. If an invitation already exists for this email, the tool returns the existing invitation details.
manage_testsManage a test (unit test) by action. For 'create', yaml_source is required (YAML string for the test definition). If the test already exists (POST 409 or 422), update it. For 'run' or 'delete', namespace and id_ are required.
manage_appsManage an app by an action (create, enable, disable, delete). For 'create', yaml_source is required (YAML string for the app definition). For 'enable', 'disable', or 'delete', uid is required.
search_appsList existing apps, optionally filtered by namespace, flowId, tags, or full-text search string.
manage_announcementsManage announcements (banners): list, create, update, or delete.
manage_groupManage a group by action. The action must be one of 'create', 'get', 'update', or 'delete'.
The tenant is managed by the Kestra client and base URL.
manage_invitationsManage an invitation by action. The action must be one of 'get' or 'delete'.
get_instance_infoGet instance metadata: configuration (default) or license info.
generate_appGenerate or regenerate an app based on a prompt and existing app context.
generate_testGenerate or regenerate a unit test suite based on a prompt and existing test context.
execute_flowStart a flow execution for a given namespace and flow_id.
manage_executionsManage an execution by action. Base don the action, the tool returns:
add_execution_labelsAdd or update labels of a terminated execution. The execution must be in one of the terminal states: SUCCESS, WARNING, FAILED, KILLED, CANCELLED, or SKIPPED. Raises a ValueError otherwise.
list_executionsList executions in a namespace (and optional flow) with flexible filtering options.
namespace_file_actionPerform a file operation in the given namespace. Returns: for 'get', 'create', 'move', 'delete', and 'search' actions, see their respective return types.
namespace_directory_actionPerform a directory operation in the given namespace.
search_flowsSearch flows by name or content. Use this tool when the user asks to:
find_flowFind a flow by approximate or partial name. Use this when you know a flow's name (or part of it) but not its exact namespace and ID.
list_flows_with_triggersList all flows that have one or more triggers. Returns a markdown‑style list of lines, e.g.:
create_flow_from_yamlCreate or update a Kestra flow from a raw YAML definition. Returns the created or updated flow's JSON metadata.
manage_flowManage a flow by action. The action must be one of 'enable', 'disable', 'list_revisions', 'list_dependencies', 'get_yaml', or 'delete'.
Returns:
generate_flowGenerate or regenerate a flow based on a prompt and existing flow context.
generate_dashboardGenerate or regenerate a dashboard based on a prompt and existing dashboard context.
manage_kv_storePerform a KV operation in the given namespace. Returns:
get_execution_logsGet logs for a specific execution with optional filtering.
download_execution_logsDownload logs for a specific execution as plain text.
search_logsSearch logs across all executions with flexible filtering options.
delete_execution_logsDelete logs for a specific execution with optional filtering.
delete_flow_logsDelete logs for all executions of a specific flow.
follow_execution_logsFollow logs for a specific execution in real-time.
list_namespacesList all existing namespaces, with an option to only include those with flows. Optionally filter by a text query. Handles pagination automatically.
list_flows_in_namespaceRetrieve all flows in a given namespace.
list_namespace_dependenciesRetrieve all flow-to-flow dependencies within a given namespace and render them as an ASCII dependency graph using only the flow IDs. Always return the legend after the graph.
replay_executionReplay (NOT RESTART, BUT REPLAY) one or more past executions by ID, or the latest execution for a given flow if a list of ids is not provided. Returns a summary string if replaying the latest execution, otherwise the API JSON response.
restart_executionRestart a failed execution. If execution_id is provided, restart that execution. If not, retrieve the latest failed execution for the given namespace and flow_id and restart it. Returns the restarted execution's JSON metadata.
change_taskrun_stateRestart the execution from a failed task or change the state of a task run. If the task_run_id is omitted, find the first failed task in that execution and mark it SUCCESS to restart downstream tasks. Returns the updated execution object.
resume_executionResume one or more paused executions by ID, or the most recent paused execution for a given namespace and flow.
If execution_ids are not provided, requires namespace and flow_id, fetches the most recent paused execution for the given flow, and resumes it.
If on_resume is provided, its key→value pairs are sent as multipart/form-data; otherwise, uses the bulk resume endpoint or fetches defaults from the Pause task if available.
force_run_executionForce run an execution only if it is in CREATED, PAUSED, or QUEUED state. Otherwise, raises a ValueError with the current execution state.
BEFORE YOU INSTALL
39 tools registered in the project's code. 10 of the 39 tools work only on Kestra Enterprise or Kestra Cloud. On the free open-source edition, set KESTRA_MCP_DISABLED_TOOLS=ee in the server's environment settings to switch them off.
INSTALLATION
docker run -i --rm --pull always -e KESTRA_BASE_URL -e KESTRA_TENANT_ID -e KESTRA_MCP_DISABLED_TOOLS -e KESTRA_MCP_LOG_LEVEL -e KESTRA_USERNAME -e KESTRA_PASSWORD ghcr.io/kestra-io/mcp-server-python:latest{
"mcpServers": {
"kestra": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"--pull",
"always",
"-e",
"KESTRA_BASE_URL",
"-e",
"KESTRA_TENANT_ID",
"-e",
"KESTRA_MCP_DISABLED_TOOLS",
"-e",
"KESTRA_MCP_LOG_LEVEL",
"-e",
"KESTRA_USERNAME",
"-e",
"KESTRA_PASSWORD",
"ghcr.io/kestra-io/mcp-server-python:latest"
],
"env": {
"KESTRA_BASE_URL": "http://host.docker.internal:8080/api/v1",
"KESTRA_TENANT_ID": "main",
"KESTRA_MCP_DISABLED_TOOLS": "ee",
"KESTRA_MCP_LOG_LEVEL": "ERROR",
"KESTRA_USERNAME": "admin@kestra.io",
"KESTRA_PASSWORD": "<YOUR_KESTRA_PASSWORD>"
}
}
}
}admin@kestra.io is the project's sample username; yours is the one you sign in to Kestra with, and <YOUR_KESTRA_PASSWORD> is your Kestra password. KESTRA_BASE_URL points at a Kestra running on this computer on port 8080; change it if yours runs elsewhere.
{
"type": "io.kestra.plugin.ai.tool.DockerMcpClient",
"image": "ghcr.io/kestra-io/mcp-server-python:latest",
"env": {
"KESTRA_BASE_URL": "{{inputs.base_url}}",
"KESTRA_TENANT_ID": "demo",
"KESTRA_USERNAME": "anna@kestra.io",
"KESTRA_PASSWORD": "{{ kv('KESTRA_PASSWORD') }}"
}
}anna@kestra.io and demo are the sample Kestra username and tenant ID. Leave the {{ … }} values as written: when the flow runs, Kestra fills in the address from a flow input named base_url and the password from a key-value entry named KESTRA_PASSWORD. Your flow needs that input, and the entry has to exist before it runs.
MCP servers
aliyun
fusedio
Lumenore-Platform
santhoshravindran7