Runpod MCP server

• Lists your Pods

  Returns the Pods in your account, newest first. Pods that belong to a cluster are left out unless you ask for them.

  Try: “Show me all my Runpod Pods and which are running.”

  Read · list-pods

• Shows one Pod in detail

  Returns a single Pod by its id, including its status, hardware and which actions are allowed right now.

  Try: “Get the details of my pod abc123, including its status.”

  Read · get-pod

• Lists your Serverless endpoints

  Returns a trimmed summary of each Serverless endpoint: name, image, compute setup, worker counts, scaling and data centers.

  Try: “Which Serverless endpoints do I have, and how many workers does each run?”

  Read · list-endpoints

• Shows one Serverless endpoint

  Returns the full record of one Serverless endpoint by its id.

  Try: “Show the full config of my endpoint my-endpoint.”

  Read · get-endpoint

• Lists your templates

  Returns each template you can see with its name, image and whether it is for Serverless, 20 per page.

  Try: “List my Runpod templates.”

  Read · list-templates

• Shows one template

  Returns the full container setup of a template, whether it is yours or from the public catalog.

  Try: “Show me the setup of my pytorch template.”

  Read · get-template

• Browses the public template catalog

  Lists templates Runpod curated, templates the community shared, or ones Runpod has verified, for both Pods and Serverless.

  Try: “Find an official Runpod template for ComfyUI.”

  Read · list-public-templates

• Lists your network volumes

  Returns the network-attached storage volumes you own, newest first.

  Try: “What network volumes do I have and how big are they?”

  Read · list-network-volumes

• Shows one network volume

  Returns one network volume by id.

  Try: “Show details for network volume nv-123.”

  Read · get-network-volume

• Lists your global volumes

  Returns your global volumes, the storage that mounts in any data center.

  Try: “List my global volumes.”

  Read · list-global-volumes

• Shows one global volume

  Returns one global volume by id.

  Try: “Show my global volume called datasets.”

  Read · get-global-volume

• Lists your Instant Clusters

  Returns the multi-Pod clusters you own, newest first.

  Try: “Do I have any Instant Clusters running?”

  Read · list-clusters

• Shows one cluster

  Returns one cluster with a summary of its Pods by status.

  Try: “Give me the status of cluster cl-1.”

  Read · get-cluster

• Lists the Pods inside a cluster

  Returns each member Pod of a cluster as a full Pod record, not just counts.

  Try: “Which Pods belong to cluster cl-1?”

  Read · list-cluster-pods

• Lists GPU types with prices

  Returns the available GPU types and their pricing, and can add live stock for a product. Can narrow to GPUs deployable in given countries.

  Try: “Which GPUs can I rent in the EU and what do they cost per hour?”

  Read · list-gpu-types

• Shows one GPU type

  Returns one GPU type with its pricing, and its stock if you ask for availability.

  Try: “How much is an NVIDIA GeForce RTX 4090 per hour?”

  Read · get-gpu-type

• Lists CPU types

  Returns the CPU flavors you can rent, with pricing, and stock when you ask for it.

  Try: “What CPU types does Runpod offer?”

  Read · list-cpu-types

• Shows one CPU type

  Returns one CPU type with pricing and optional availability.

  Try: “Show pricing for the cpu3c CPU type.”

  Read · get-cpu-type

• Lists Runpod data centers

  Returns each location with region, compliance and supported storage tiers, and can add GPU or CPU stock per location.

  Try: “Which data centers have H100s in stock right now?”

  Read · list-data-centers

• Shows one data center

  Returns one data center, with GPU or CPU stock if you ask for it.

  Try: “Show the EU-RO-1 data center and its GPU stock.”

  Read · get-data-center

• Shows GPU stock by CUDA version

  Builds a matrix of GPU stock across host CUDA versions, or probes chosen CUDA versions for graded stock and lowest price. Helps tell a stock problem from a compatibility problem.

  Try: “Why does my endpoint have no workers? Check GPU capacity for CUDA 12.8.”

  Read · get-capacity

• Shows your spend over time

  Returns total spend grouped into time buckets, split into Pods, Serverless and storage.

  Try: “How much did I spend on Runpod over the last 7 days?”

  Read · list-billing

• Shows Pod billing

  Returns Pod-only charges per time bucket, for all Pods or one Pod.

  Try: “What did my pod abc123 cost this week?”

  Read · list-pod-billing

• Shows Serverless billing

  Returns Serverless charges per time bucket. With many endpoints the default output is huge, so ask for the last bucket only.

  Try: “Total Serverless spend for the last month, one bucket.”

  Read · list-serverless-billing

• Shows storage billing

  Returns network volume charges per time bucket, for all volumes or one.

  Try: “What does my network storage cost per day?”

  Read · list-network-volume-billing

• Shows cluster billing

  Returns Instant Cluster charges per time bucket, for all clusters or one.

  Try: “What did cluster cl-1 cost this week?”

  Read · list-cluster-billing

• Shows Public Endpoint billing

  Returns what you spent on Runpod Public Endpoints per time bucket.

  Try: “How much did I spend on public endpoint calls this month?”

  Read · list-endpoint-billing

• Reads the account audit trail

  Returns who did what to which resource in your account, newest first, with filters and paging.

  Try: “Who deleted a network volume in the last 3 days?”

  Read · list-audit-logs

• Shows your organization

  Returns the organization you currently belong to.

  Try: “What organization am I in?”

  Read · get-organization

• Lists organization members

  Returns every member of your organization with their role.

  Try: “Who is in my Runpod organization?”

  Read · list-organization-users

• Shows one organization member

  Returns one member by user id with the role they hold.

  Try: “What role does user u-12 have?”

  Read · get-organization-user

• Lists your registered SSH keys

  Returns the public SSH keys on your account, which are put into Pods that start with SSH.

  Try: “Which SSH keys are registered on my account?”

  Read · get-ssh-keys

• Lists your secrets

  Returns the names and metadata of your account secrets. Values are never returned.

  Try: “What secrets do I have stored on Runpod?”

  Read · list-secrets

• Shows one secret

  Returns one secret record by id. The value is write-only and is never returned.

  Try: “Show the details of my HF_TOKEN secret.”

  Read · get-secret

• Lists container registry logins

  Returns the private container registry credentials you saved, without usernames or passwords.

  Try: “Which container registries have I connected?”

  Read · list-registries

• Shows one registry login

  Returns one saved registry credential record, without the username or password.

  Try: “Show my docker hub registry entry.”

  Read · get-registry

• Lists ECR delegations

  Returns the AWS ECR delegations registered for your account.

  Try: “List my ECR delegations.”

  Read · list-delegations

• Lists workers on an endpoint

  Returns the active workers of a Serverless endpoint with their state, plus counts by state. Best for finding out why a job is stuck.

  Try: “Why is my endpoint stuck? List its workers and their states.”

  Read · list-endpoint-workers

• Shows endpoint release history

  Returns the endpoint release history, newest first, each with a diff of what changed and how many workers run the current version.

  Try: “What changed in the last release of my endpoint?”

  Read · list-endpoint-releases

• Lists endpoint GitHub builds

  Returns the endpoint GitHub build history, newest first.

  Try: “Did the last build of my endpoint succeed?”

  Read · list-endpoint-builds

• Shows one endpoint build

  Returns one GitHub build of an endpoint by id, however old it is.

  Try: “Show build b-77 for my endpoint.”

  Read · get-endpoint-build

• Shows endpoint health

  Returns worker counts by state and job queue numbers for a Serverless endpoint. The tool notes this view can lag, so list the workers when a job is stuck.

  Try: “Is my endpoint healthy? How many jobs are queued?”

  Read · endpoint-health

• Checks a Serverless job

  Returns the status and output of a job. It can wait on the server until the job finishes, which avoids repeated polling through a cold start.

  Try: “Check job j-123 and wait up to a minute for the result.”

  Read · get-job-status

• Collects streamed job output

  Reads streamed output from a job until it ends or a time budget runs out, then returns the chunks so far. The worker must support streaming.

  Try: “Stream the output of job j-123.”

  Read · stream-job

• Reads a Pod's logs

  Returns a short snapshot of a Pod's container or system logs as parsed lines. A flag says when the output was cut at the size cap.

  Try: “Show me the last logs from pod abc123.”

  Read · stream-pod-logs

• Reads a worker's logs

  Returns a short snapshot of a Serverless worker's logs. Use it when a job stays in the queue while workers exist.

  Try: “Show logs for worker w-9 on my endpoint.”

  Read · stream-worker-logs

• Browses Runpod Public Endpoints

  Lists ready-made, pay-per-use model APIs for text, image, video and audio that need no deployment. The catalog is public.

  Try: “Which Runpod Public Endpoints can generate images?”

  Read · list-public-endpoints

• Browses the Runpod Hub

  Lists prebuilt Serverless workers and Pod templates such as vLLM and ComfyUI, most deployed first. Can include each release config.

  Try: “What are the most popular repos on the Runpod Hub?”

  Read · list-hub-repos

• Lists Runpod guides

  Lists the official guides the server bundles: tool skills, reference docs and step-by-step walkthroughs.

  Try: “What Runpod guides are available?”

  Read · list-guides

• Reads one Runpod guide

  Returns a whole guide by its id from the list of guides.

  Try: “Read the Runpod guide on dev pods.”

  Read · read-guide

• Looks up a Runpod concept

  Returns what a concept like pod, worker or network volume is, with its rules and the public source for each.

  Try: “What does a stopped pod keep on Runpod?”

  Read · lookup-concept

• Searches Runpod rules

  Finds the rules in the concept set that match your words and returns them with their sources.

  Try: “Search for rules about network volumes and data centers.”

  Read · search-concepts

• Shows a tool's full argument list

  Returns one tool's complete definition. Some tools list only their basic shape, and the assistant calls this before using them.

  Try: “Describe the create-endpoint tool.”

  Read · describe-tool

• Searches Runpod docs

  Searches Runpod documentation, support knowledge and blog and returns passages with links. The query goes to the hosted Runpod server, even when you run the server locally.

  Try: “How do I attach a network volume to a Serverless endpoint?”

  Network · search-docs

• Rents a new Pod

  Creates a GPU or CPU Pod. It bills by the hour from the moment it is created until you stop or terminate it.

  Try: “Create a pod named test-pod with one RTX 4090 and the pytorch image.”

  Write · create-pod

• Changes a Pod's settings

  Changes only the fields you send on a Pod. Sending an empty value, such as an empty ports list, clears that field.

  Try: “Change the container disk of pod abc123 to 50 GB.”

  Delete · update-pod

• Starts, stops, restarts or terminates a Pod

  Moves a Pod to a new state with one action. Terminate deletes the Pod for good, and stop or restart wipe the container disk.

  Try: “Stop pod abc123.”

  Delete · pod-action

• Terminates a Pod permanently

  Deletes a Pod and loses its container disk. The tool text says to check the id first.

  Try: “Terminate pod abc123.”

  Delete · delete-pod

• Creates a Serverless endpoint

  Creates an endpoint from an image and a GPU pool or CPU setup. It is billable, and the tool text says to quote the hourly price first.

  Try: “Create a Serverless endpoint named my-endpoint on AMPERE_80, 0 min and 3 max workers.”

  Write · create-endpoint

• Changes a Serverless endpoint

  Changes only the fields you send. Sending GPU pools without exclusions clears exclusions set elsewhere.

  Try: “Raise max workers on my-endpoint to 5.”

  Delete · update-endpoint

• Deletes an endpoint

  Removes a Serverless endpoint permanently and immediately.

  Try: “Delete the endpoint test-endpoint.”

  Delete · delete-endpoint

• Sets which GPUs an endpoint uses

  Replaces the GPU choice of a Serverless endpoint, including pinning one GPU type, and leaves its other settings alone. Exclusions you do not resend are cleared.

  Try: “Pin my-endpoint to NVIDIA L40S workers only.”

  Delete · set-endpoint-gpus

• Submits a job to an endpoint

  Sends a job to a Serverless endpoint and returns a job id right away. Results stay available for 30 minutes after the job completes.

  Try: “Run my-endpoint with the prompt "a red fox in snow".”

  Write · run-endpoint

• Runs a job and waits for the result

  Sends a job and waits for the answer. It suits warm endpoints and quick jobs, and falls back to a job id when the job takes longer.

  Try: “Run a quick test on my-endpoint and show me the output.”

  Write · runsync-endpoint

• Cancels a job

  Stops a Serverless job that is queued or running.

  Try: “Cancel job j-123.”

  Delete · cancel-job

• Retries a failed job

  Requeues a job that failed or timed out. Its earlier output is removed.

  Try: “Retry job j-123.”

  Delete · retry-job

• Clears an endpoint's waiting jobs

  Removes all pending jobs from an endpoint queue. Jobs already running continue.

  Try: “Purge the queue on my-endpoint.”

  Delete · purge-endpoint-queue

• Deploys a Hub repo as an endpoint

  Creates a new Serverless endpoint from a Hub release, as if you had clicked Deploy on the Hub. It uses Runpod's GraphQL API.

  Try: “Deploy runpod-workers/worker-comfyui from the Hub.”

  Write · deploy-hub-repo

• Creates a template

  Saves a reusable container setup of image, disk, ports, env and mounts that Pods and endpoints can use.

  Try: “Create a template called my-vllm from the vllm image.”

  Write · create-template

• Changes a template

  Changes only the fields you send on a template you own.

  Try: “Add port 8888 to my-vllm.”

  Delete · update-template

• Deletes a template

  Deletes a template permanently. The tool text warns against deleting one a live endpoint uses.

  Try: “Delete my old template pytorch-old.”

  Delete · delete-template

• Creates a network volume

  Creates persistent storage with a name, size in GB and data center. It is billable.

  Try: “Create a 100 GB network volume called models in EU-RO-1.”

  Write · create-network-volume

• Changes a network volume

  Updates a network volume's name or size. The size can only grow.

  Try: “Grow volume models to 200 GB.”

  Write · update-network-volume

• Deletes a network volume

  Deletes a network volume and its storage.

  Try: “Delete network volume nv-123.”

  Delete · delete-network-volume

• Creates a global volume

  Creates storage that mounts in any data center. Accounts have a cap on how many they can hold.

  Try: “Create a global volume called datasets.”

  Write · create-global-volume

• Changes a global volume

  Updates the fields you send on a global volume.

  Try: “Rename global volume datasets to training-data.”

  Write · update-global-volume

• Deletes a global volume

  Deletes a global volume and its data. It fails while the volume is still mounted.

  Try: “Delete global volume gv-1.”

  Delete · delete-global-volume

• Creates an Instant Cluster

  Rents several identical Pods with fast interconnect at once, all billable from creation. The tool text says to quote the total hourly price and wait for your go.

  Try: “Create a cluster of 2 pods with 8 H100s each.”

  Write · create-cluster

• Renames a cluster

  Changes only the name of a cluster. Its size and setup are fixed at creation.

  Try: “Rename cluster cl-1 to training.”

  Write · update-cluster

• Deletes a cluster

  Deletes a cluster and terminates every Pod in it.

  Try: “Delete cluster cl-1.”

  Delete · delete-cluster

• Stores a secret

  Saves an encrypted value under a unique name. The value cannot be read back, and Pods can use it through an environment variable placeholder.

  Try: “Store my Hugging Face token as a secret named HF_TOKEN.”

  Write · create-secret

• Rotates a secret

  Changes a secret's value or description. Running Pods keep the old value until their next boot.

  Try: “Rotate the value of HF_TOKEN.”

  Delete · update-secret

• Deletes a secret

  Deletes a secret. Environment variables that use it stop resolving at the next Pod or worker boot.

  Try: “Delete secret HF_TOKEN.”

  Delete · delete-secret

• Saves a registry login

  Stores credentials for a private container registry. They are write-only.

  Try: “Add my GitHub Container Registry login.”

  Write · create-registry

• Deletes a registry login

  Deletes saved registry credentials. It is refused if a Pod uses them, but templates that use them silently lose the link.

  Try: “Delete my old docker registry login.”

  Delete · delete-registry

• Registers an ECR delegation

  Registers an AWS ECR delegation for your account.

  Try: “Register an ECR delegation for my AWS account.”

  Write · create-delegation

• Revokes an ECR delegation

  Revokes a registered ECR delegation.

  Try: “Revoke ECR delegation d-1.”

  Delete · revoke-delegation

• Replaces your SSH keys

  Replaces the whole set of public SSH keys on your account. Any key you leave out is removed, and an empty list removes all.

  Try: “Add my new public key to my Runpod SSH keys.”

  Delete · update-ssh-keys

Watch out for

• It can delete Pods, endpoints and volumes

  Several tools delete or terminate resources for good, including delete-pod, delete-endpoint, delete-cluster and delete-network-volume. Set your client to ask before each call. github.com

• It can start billable GPUs

  create-pod, create-endpoint, create-cluster and create-network-volume start charges that run until you stop or delete them. The server text tells the assistant to quote the price first, but nothing in the code stops a call. github.com

• It acts with your full account rights

  The README says the server acts with the full permissions of the key or sign-in. There is no read-only switch. github.com

• It tells the assistant to load playbooks first

  The server instructions say to read a routed guide before the first Runpod call and call a reply without it out of contract. Expect extra guide reads at the start of a session.

• The hosted server records anonymous tool use

  It logs tool name, status, duration and transport for each call, and drops arguments except on guide and concept tools. Send the X-Runpod-Analytics: off header to opt out. The local server sends none of this, but docs search still goes to the hosted server. github.com

Security advisories