Skip to main content
Mission Control MCP is in public preview. Tool coverage and behavior may change while the service is in preview. To share feedback or report an issue, email support@coreweave.com.
Mission Control MCP is a hosted Model Context Protocol (MCP) server for CoreWeave infrastructure. It connects MCP-compatible clients, such as Claude Code, Cursor, and Codex, to read-only tools for observing, triaging, and optimizing CoreWeave environments. With a CoreWeave API token, you can ask natural-language questions about your infrastructure and let an agent gather current evidence from metrics, logs, dashboards, documentation, cluster state, node details, and object storage context. The agent works inside the same visibility boundary as your token, so it can access what your token can access. The server is hosted by CoreWeave, so there is nothing to deploy or operate. You point your client at the endpoint and authenticate with a token. Because the tools are read-only, an agent can observe and triage your infrastructure but cannot change it.

​
What you can do

Mission Control MCP brings several CoreWeave tool families into one MCP connection. The tools are small and composable, so an agent can build a multi-step investigation instead of relying on one opaque “debug everything” call. It does not replace dashboards, CLIs, or documentation. It makes those interfaces easier for an agent to use, and easier for you to reason over, during real operational work.
  • Observability: query metrics, search logs, inspect dashboards and the queries behind their panels, and discover datasources, backed by Grafana-compatible systems.
  • CoreWeave documentation: search and read CoreWeave public documentation so an agent answers product and API questions from documentation rather than model memory.
  • CoreWeave resources: inspect clusters, nodes, CoreWeave SUNK resources, and CoreWeave AI Object Storage buckets, then connect that state to observability data.
For the full catalog, including each tool’s inputs and outputs, see the Mission Control MCP tool reference.

​
Who should use it

It is useful across a range of roles, from ML engineers debugging training runs to platform teams assembling an evidence chain.
  • AI researchers and ML engineers can debug training and inference workloads without leaving their editor, asking for GPU health signals, node status, workload logs, or relevant dashboards and following up with further questions.
  • Platform and infrastructure teams can assemble an evidence chain quickly: what is affected, which metrics changed, which errors appear in the logs, and which dashboards are relevant.
  • Application teams and operators can answer operational questions that cross product surfaces, such as combining workload logs, metrics, and object storage state to investigate performance issues in training jobs.
  • Domain experts can supply the framing while the agent handles tool choreography, turning operational telemetry into tables, prototypes, and concrete next steps.

​
Set up Mission Control MCP

To connect a client and start calling the tools, see Configure Mission Control MCP. Setup requires a CoreWeave API token and an MCP-compatible client, such as Claude Code, Cursor, or Codex.

Configure Mission Control MCP

Connect Claude Code, Cursor, or another MCP client to the hosted server.

Tool reference

Every Mission Control MCP tool, grouped by family, with inputs and outputs.

CoreWeave Grafana

Managed Grafana dashboards, metrics, and logs that Mission Control MCP tools observe.

CoreWeave AI Object Storage

S3-compatible object storage that the object storage tools inspect.
Last modified on June 25, 2026