Get startedDevelopers · MCP access
Connect agents with controlled execution
Connect an MCP-compatible agent to workspace-scoped datasets, models, evaluations, and deployments.
Source reviewed 2026-09-16. Availability depends on your installation, permissions, and compatible runtimes. A supported path is not a guarantee of model quality or production readiness.
Outcome
Use read scopes for inspection, or separately grant workflow request and cancellation scopes. Requested execution waits for human approval before compute.
Start here
What you need to know first
Check the inputs below before starting. If you are new, begin with the first-project guide. A dataset holds media and labels; a model produces results; a deployment makes a selected model version callable. Creating one does not create the others.
Bring these inputs
- — The MCP endpoint configured for this installation
- — A workspace API key with the necessary per-tool scopes
Expected output and limits
- — Structured dataset, model, evaluation, and deployment evidence
- — Stable workspace identifiers an agent can reference in its answer
- — An approval-gated workflow request when execution scope is granted
Core workflow
Choose a client and its access
In Build & evaluate → Agents, choose Claude Code, Cursor, or another HTTP MCP client. Use Explore only for read access, explicitly allow workflow requests, or use an existing key. Installation details appear after the key is ready. Only an owner or admin can create a key here; existing-key permissions stay unchanged.
Connect the MCP endpoint
Set SCORE_API_KEY in your client's environment and follow the visible client-specific setup. Check the server's tools in your client; key creation is not a connection test. The native local route is http://localhost:8000/v1/mcp. The target public endpoint https://mcp.scorestudio.ai/mcp requires separate deployment acceptance.
Inspect or request a run
List datasets, models, evaluations and deployments. request_workflow_run needs an active saved workflow and an idempotency key; it returns a paused request and human review link. Inspect with get_workflow_run.
Rotate or revoke access
Use Manage and revoke keys to open Settings → API keys. Replace the credential when its scope changes or remove it when the agent no longer needs access. New keys cannot be retrieved after leaving the page.
What this surface supports
Remote MCP endpoint
Dataset and model discovery
Evaluation evidence
Deployment inspection and approval-gated workflow requests
Expert section
Contracts, signals, and failure modes
Use this section when you are defining acceptance criteria, automating the surface, or reviewing whether its output is safe to promote downstream.
Quality and operating signals
- — MCP initialization and tool availability
- — Per-tool authorization
- — Version and deployment state returned by Score Studio
Common failure modes
- — Exposing an API key in source control or chat history
- — Missing scopes for a requested evidence type
- — Treating read access as permission to release or mutate a system
Expert release checklist
□ Inputs and dependencies are pinned to immutable versions.
□ Acceptance metrics include critical classes and operating slices.
□ Failure, retry, cost, and rollback behavior are understood.
□ The resulting artifact has an owner and a downstream review path.
Engineering safeguards
- — Read-only tools do not change workspace state
- — Every query remains organization-scoped
- — API-key scopes are enforced per tool
- — Agents cannot approve their own requests; cancellation may leave independent jobs running