Rayrun

Rayrun is a hosted gateway for bringing multiple Model Context Protocol tools and services behind one endpoint for coding agents. Instead of placing every server URL and credential in each client configuration, a workspace connects the services once and gives clients a single Rayrun URL. The platform holds credentials, discovers tool definitions, applies access policy to calls, and records activity. It works with clients named on the site including Claude, Codex, Cursor, VS Code, Windsurf, and OpenCode. The result is a central control layer for developers or teams that want agents to use external systems without scattering secrets and policy across every local setup.

A workspace can connect a remote MCP server, import an OpenAPI document, host an npm package or container image, or deploy a custom MCP server from source. Rayrun turns supported OpenAPI operations into typed tools and completes OAuth when a remote service needs authorization. For a service that does not yet exist, its command-line workflow scaffolds a TypeScript or Python project for a coding agent to extend. Deployment uploads the source, runs validation and an isolated build, verifies a candidate release, and only then moves traffic. If the build fails, the current release remains live. This gives teams a route from an agent-written server to a managed endpoint without assembling separate hosting infrastructure. Hosted package and image services can use optional persistent data storage, and inactive source images remain available long enough to support rollback.

Governance is central to the product rather than an optional reporting layer. Administrators can choose allow, ask, or block behavior per service and override the setting for an individual tool. Rayrun can make a decision using the client, tool, and arguments, while exact calls that require a person wait for approval. Newly discovered definitions are recorded, and later definition changes can be held for review before agents see them. The site also describes content scanning in both directions for secrets, personal data, and prompt injection. Every invocation leaves an activity record with details such as the user, client, service, policy decision, duration, and result, making agent actions easier to inspect after the fact. A workspace-wide daily safety budget can stop new upstream calls at a chosen limit independently of the monthly billing allowance.

Credential handling follows the same centralized model. Stored credentials remain with Rayrun or can be kept in the customer's own AWS account, while clients receive a token and upstream services receive the needed credential. The homepage says stored secrets are not sent back to a browser, even to the administrator who entered them. Team features add a shared credential vault, roles, client-specific access, and pooled call usage. Enterprise options include single sign-on through OIDC or SAML, SCIM provisioning, role-based access, and signed batched audit streaming to a SIEM. Rayrun publishes a security overview that describes its controls and the certification areas that are not yet claimed, which gives security reviewers concrete material to examine. Each workspace also has its own data-encryption key according to the published overview.

The platform is aimed at developers building agent workflows and teams responsible for governing them. A solo developer can use it to avoid repeating MCP configuration across several coding clients. A platform or security team can use the shared endpoint to control which agent may call which service, review definition changes, and trace activity across tools. Teams can also publish versioned workspace Skills through MCP, add hosted TypeScript hooks, or automate management through the public API, SDKs, and webhooks listed in the documentation. Code mode reduces the need to load every upstream tool definition at once by retrieving a definition when the model needs it.

Rayrun uses a freemium model. The free tier is for one developer and includes the core gateway, hosting, governance, credential, hook, and audit features. It allows up to 25 source-built services, 50 isolated builds in a 24-hour period, 5 GiB of retained images, and 50,000 tool calls per month. The Team plan costs $25 per developer each month and adds the shared vault, roles, client-specific access, and 100,000 pooled calls per developer. Usage above the allowance costs $0.20 per thousand calls, while refused calls and tool discovery do not count. Enterprise service starts at $1,500 monthly. This structure makes the free plan substantial for individual use, but shared governance and identity controls move teams into paid tiers.