Integration

Runpod integration

Runpod rents GPU and CPU cloud compute. Create, update, start, stop and terminate Pods, deploy and scale Serverless endpoints, run Instant Clusters, manage templates, network and global volumes, container registry credentials, account secrets (values are write-only) and SSH public keys, browse the GPU, CPU and data center catalog with live availability, and read billing history and audit logs (REST API v2). Live log streams (Server-Sent Events) are not exposed, and environment variables and container command lines (args, cmd, entrypoint) are left out of pod, endpoint and template answers because they may hold secrets. Authentication is a Runpod API key sent as a Bearer token.

Connect Runpod How integrations work

Ask it in chat

Say what should happen.
The workflow builds itself.

When this happens, do that in Runpod.Built from the 65 Runpod tools below, on the connection you add once.

What you can do with Runpod

Each of these is one operation on one endpoint, available both as a workflow step and as a tool an AI agent can call on its own. Nothing bundles several operations into one action, so what a run did stays readable afterwards.

Most endpoints run on the free plan. A few, mainly the ones that publish on your behalf, need a paid plan.

Pod, 7 operations

create_podCreate podPOSTCreates a GPU or CPU pod on the live account, billed while it runs. Send name, exactly one of gpu or cpu, and image unless templateId is set. It starts PROVISIONING: poll get_pod until status is RUNNING.
delete_podDelete podDELETEPermanently terminates a pod (irreversible), ending its billing: compute is released and host-local persistent storage is destroyed; network volumes are only detached. Pods in a cluster are removed with delete_cluster.
get_podGet podGETReturns a single pod by ID.
list_cluster_podsList cluster podsGETLists every member pod of an Instant Cluster as a full pod object (get_cluster only gives counts).
list_pod_billingList pod billingGETGets pod billing per pod per time bucket (GPU, CPU, disk, total); podId narrows to one pod. Use startTime/endTime or lastN with bucketSize.
list_podsList podsGETLists your pods, newest first (cluster member pods only with includeClusterPods=true). Page with cursor = pagination.nextCursor while pagination.hasNextPage is true.
update_podUpdate podPATCHChanges a live pod's settings; send only fields to change (name, image, args, disk, ports, env, registry, mounts, locked, globalNetworking). Some apply on next restart. Cluster pods cannot be updated.

Cluster, 6 operations

create_clusterCreate clusterPOSTCreates a multi-node Instant Cluster on the live account, billed while it runs: compute sets podCount identical GPU pods; give image or templateId for their container.
delete_clusterDelete clusterDELETEPermanently deletes an Instant Cluster and terminates all of its member pods (irreversible), ending their billing.
get_clusterGet clusterGETReturns a single cluster by ID. The pods field is an aggregate summary (total + count by status); fetch the member pods themselves from list_cluster_pods.
list_cluster_billingList cluster billingGETGets Instant Cluster billing per cluster per time bucket (GPU, disk, networking, total); clusterId narrows to one cluster.
list_clustersList clustersGETLists your Instant Clusters, newest first. Page with cursor = pagination.nextCursor while pagination.hasNextPage is true.
update_clusterUpdate clusterPATCHRenames a cluster. This endpoint only changes the cluster name, compute shape, type, and container configuration are fixed at creation and cannot be updated.

Network volume, 6 operations

create_network_volumeCreate network volumePOSTCreates a network volume (persistent storage mountable into pods and workers) on the live account. It is billed for storage until deleted. Send name, size in GB and dataCenter; the tier (type) cannot change later.
delete_network_volumeDelete network volumeDELETEPermanently deletes a network volume and releases its storage (its data is lost).
get_network_volumeGet network volumeGETReturns a single network volume by ID.
list_network_volume_billingList network volume billingGETGets network volume billing per volume per time bucket (standard and high-performance storage); networkVolumeId narrows to one volume.
list_network_volumesList network volumesGETLists your network volumes, newest first. Page with cursor = pagination.nextCursor while pagination.hasNextPage is true.
update_network_volumeUpdate network volumePATCHRenames or enlarges a network volume; send only fields to change. size can only grow, never shrink.

Serverless, 6 operations

create_endpointCreate endpointPOSTCreates a Serverless endpoint on the live account. Send name, type, scaling, exactly one of gpu or cpu, and image unless templateId is set. No worker runs (or bills) until a job arrives unless workers.min > 0.
delete_endpointDelete endpointDELETEPermanently deletes a Serverless endpoint and its bound template (irreversible), ending its spend: all workers are terminated and queued or running jobs are cancelled.
get_endpointGet endpointGETReturns a single serverless endpoint by ID.
list_endpointsList endpointsGETLists your Serverless endpoints, newest first, with their job request URLs. Page with cursor = pagination.nextCursor while pagination.hasNextPage is true.
list_serverless_billingList serverless billingGETGets Serverless endpoint billing per endpoint per time bucket (GPU, CPU, disk, fee, total); serverlessId narrows to one endpoint.
update_endpointUpdate endpointPATCHChanges a live Serverless endpoint; send only fields to change. Scaling and worker bounds apply promptly, and raising workers.min keeps that many workers running and billed. Image or env changes roll out as a new release (list_endpoint_releases).

Template, 6 operations

create_templateCreate templatePOSTCreates a reusable container configuration (image, disk, ports, env, registry, mounts) for pods or Serverless endpoints; pass its id as templateId to create_pod or create_endpoint. public=true shares it with every user.
delete_templateDelete templateDELETEPermanently deletes a template you own. Refused while a pod references it or a Serverless endpoint is bound to it; public catalog templates answer 404.
get_templateGet templateGETGets the full configuration of a template you own or of a public catalog template.
list_public_templatesList public templatesGETLists public templates (at most 100, no paging): source=official (default), verified or community. The serverless flag tells pod and endpoint templates apart.
list_templatesList templatesGETLists your templates (including team ones). Page with cursor = pagination.nextCursor while pagination.hasNextPage is true.
update_templateUpdate templatePATCHChanges a template you own; send only fields to change. Pods and endpoints already created from it are not changed. Public catalog templates answer 404.

Global volume, 5 operations

create_global_volumeCreate global volumePOSTCreates a global volume (storage mountable in any data center) on the live account; only name is required. An account holds a capped number of global volumes; past it the call answers 400.
delete_global_volumeDelete global volumeDELETEPermanently deletes a global volume and its data. Answers 409 while it is mounted: terminate every pod using it (stopped pods count) and detach it from every Serverless endpoint first.
get_global_volumeGet global volumeGETReturns a single global volume by ID.
list_global_volumesList global volumesGETLists your global volumes, newest first. Page with cursor = pagination.nextCursor while pagination.hasNextPage is true.
update_global_volumeUpdate global volumePATCHRenames a global volume.

Secret, 5 operations

create_secretCreate secretPOSTStores a new account secret (write-only value) on the live account; name is unique and immutable, 409 if taken. Reference it in env as {{ RUNPOD_SECRET_<name> }}.
delete_secretDelete secretDELETEPermanently deletes an account secret; environment variables that reference it no longer resolve.
get_secretGet secretGETReturns a single secret's metadata by ID. The value is write-only and never returned.
list_secretsList secretsGETLists your account secrets' names and dates (values are never returned). Reference one in env as {{ RUNPOD_SECRET_<name> }}.
update_secretUpdate secretPATCHRotates a secret's value and/or changes its description (at least one). Running pods and workers keep the old value until their next boot.

Registry, 4 operations

create_registryCreate registryPOSTStores a username and password for a private container registry; pass the returned id as registry when creating pods, endpoints or templates. The credentials are never returned.
delete_registryDelete registryDELETEPermanently deletes a container registry credential. Refused while a pod pulls with it; templates that reference it silently lose the reference instead.
get_registryGet registryGETGets a container registry credential's id and name. The username and password are never returned.
list_registriesList registriesGETReturns all container registry credentials owned by the authenticated user.

Delegation, 3 operations

create_delegationCreate delegationPOSTRegisters an Amazon ECR delegation so Runpod can pull images from an ECR repository you grant it. resource is the repository ARN.
list_delegationsList delegationsGETLists your Amazon ECR delegations (Runpod pulling images from your ECR repository), with repository, tag and region.
revoke_delegationRevoke delegationDELETERevokes an Amazon ECR delegation: Runpod can no longer pull images from that repository.

Build, 2 operations

get_endpoint_buildGet endpoint buildGETReturns one of the endpoint's GitHub builds by id, regardless of age, no need to page through `list_endpoint_builds` to reach it.
list_endpoint_buildsList endpoint buildsGETLists a Serverless endpoint's GitHub builds, newest first (100 per page by default). Page with cursor = pagination.nextCursor; fetch one with get_endpoint_build.

Cpus, 2 operations

get_cpu_typeGet cpu typeGETGets one CPU flavor with per-vCPU prices. Add include=AVAILABILITY with product (POD or SERVERLESS) for live stock.
list_cpu_typesList cpu typesGETLists CPU flavors with per-vCPU prices. Add include=AVAILABILITY with product (POD or SERVERLESS) for live stock.

Datacenter, 2 operations

get_data_centerGet data centerGETReturns a single data center. Availability is included only when requested with include=GPU_AVAILABILITY or include=CPU_AVAILABILITY.
list_data_centersList data centersGETLists data centers with region, compliance, volume tiers and global networking. include=GPU_AVAILABILITY or CPU_AVAILABILITY adds stock; regions, compliance and globalNetwork filter.

Gpus, 2 operations

get_gpu_typeGet gpu typeGETGets one GPU type with prices. Add include=AVAILABILITY with product (POD, CLUSTER or SERVERLESS) for live stock per data center.
list_gpu_typesList gpu typesGETLists GPU types with prices, per-cloud maximum counts and Serverless pool ids. Add include=AVAILABILITY with product (POD, CLUSTER or SERVERLESS) for live stock; countryCodes narrows by country.

Ssh key, 2 operations

get_ssh_keysGet ssh keysGETLists the SSH public keys registered on your account, which pods created with startSsh receive.
update_ssh_keysUpdate ssh keysPUTReplaces your account's whole set of SSH public keys (keys not sent are removed). Only pods created afterwards with startSsh get the new keys.

Action, 1 operation

pod_actionPod actionPOSTStarts, stops, restarts or terminates (permanently deletes) a live pod. Start resumes compute billing; a stopped pod is billed only for its volume disk; terminate ends billing. Allowed actions follow its actions field; terminate returns no body.

Audit log, 1 operation

list_audit_logsList audit logsGETLists the account's (or organization's) audit trail, newest first: who did which action on which resource. Answers 403 when the plan has no audit logs. Page with cursor = pagination.nextCursor.

Billing, 1 operation

list_billingList billingGETGets total spend per time bucket across all Runpod resources, with pod, Serverless, storage, endpoint and cluster components. Use startTime/endTime or lastN with bucketSize.

Endpoint, 1 operation

list_endpoint_billingList endpoint billingGETGets Runpod public endpoint billing per time bucket (not your own Serverless endpoints, see list_serverless_billing).

Organization, 1 operation

get_organizationGet organizationGETGets the id and name of the organization the API key acts for. Answers 404 when the caller belongs to no organization.

Release, 1 operation

list_endpoint_releasesList endpoint releasesGETLists a Serverless endpoint's configuration releases, newest first, each with a diff, plus how many workers run the latest version. Page with cursor.

Worker, 1 operation

list_endpoint_workersList endpoint workersGETLists the currently active workers of a Serverless endpoint with their status (RUNNING, IDLE, INITIALIZING, THROTTLED, UNHEALTHY) and a count per status. Scaled-down workers are not returned.

Use Runpod from Claude Code, Cursor or Codex

Connect Runpod in LiveContext, create an API key in Settings, then add the LiveContext MCP server to your client. Its Runpod operations become tools the client can call, next to every other integration you connected.

Claude Code

claude mcp add --transport http livecontext https://livecontext.ai/mcp --header "X-API-Key: YOUR_API_KEY"

Codex (~/.codex/config.toml)

[mcp_servers.livecontext]
url = "https://livecontext.ai/mcp"

[mcp_servers.livecontext.http_headers]
"X-API-Key" = "YOUR_API_KEY"

Cursor, Claude Desktop and other JSON configs

{
  "mcpServers": {
    "livecontext": {
      "type": "http",
      "url": "https://livecontext.ai/mcp",
      "headers": {
        "X-API-Key": "YOUR_API_KEY"
      }
    }
  }
}

Set up the MCP server step by step

How to connect Runpod

Create your workspace
Sign up free. Nothing to install.
Connect Runpod once
Add your Runpod connection in Settings. Every tool from this integration reuses it, and secrets are encrypted at rest and visible only to you.
Describe the job in chat
The workflow builds itself in front of you and picks the endpoints it needs. Drag the steps yourself on the canvas whenever you would rather be explicit.