Remote MCP

inis.run exposes a hosted remote MCP server — no local binary, no subprocess. Point your MCP client at our endpoint with your API key and the sandbox tools appear automatically.

Transport is SSE (not streamable HTTP): connect to /sse.

Org endpoint

https://mcp.inis.run/sse

Authenticate with your API key:

Authorization: Bearer inis_...

Configure your MCP client

{
  "mcpServers": {
    "inis": {
      "url": "https://mcp.inis.run/sse",
      "headers": {
        "Authorization": "Bearer inis_..."
      }
    }
  }
}

import { streamText } from "ai";
import { experimental_createMCPClient as createMCPClient } from "ai";

const mcp = await createMCPClient({
  transport: {
    type: "sse",
    url: "https://mcp.inis.run/sse",
    headers: { Authorization: `Bearer ${process.env.INIS_API_KEY}` },
  },
});

const tools = await mcp.tools();

const result = streamText({
  model,
  tools,
  maxSteps: 5,
  // ...
});

Works anywhere your agent runs — Vercel, Fly, a container, CI — because nothing needs to be installed on the host.

Org tools

Group	Tools
Execution	`execute_code`, `run_command`, `batch_exec`
Sessions	`create_session`, `list_sessions`, `get_session`, `destroy_session`, `pause_session`, `resume_session`, `fork_session`, `branch_session`
Files	`read_file`, `write_file`, `write_files`, `list_files`, `find_files`, `grep_files`, `mkdir`, `remove`, `rename`
Networking	`expose_port`, `unexpose_port`, `get_egress`, `set_egress`
Connections	`list_connections`, `revoke_connection`
Checkpoints	`create_checkpoint`, `list_checkpoints`, `get_checkpoint`, `delete_checkpoint`, `restore_session`
Artifacts	`capture_artifacts`, `list_session_artifacts`, `get_artifact`, `extend_artifact`, `delete_artifact`
Templates	`list_templates`, `save_as_template`, `delete_template`

create_session also takes an optional connections array — the same shape as the HTTP API's connections[] field (see Connections) — to attach session-scoped credential leases at create time. list_connections and revoke_connection are read-only inspection and revoke; there is no rotate/replace tool on either MCP server, org or session — rotating a Connection takes a plaintext secret as input, and an MCP tool call's arguments are typically logged by the client and echoed into the agent's own transcript, which isn't a safe place for a credential to pass through a second time. Rotate from the CLI or console instead, or create a new session with a fresh Connection.

Session MCP

For an agent bound to one session (e.g. after you create from a template), use:

https://mcp.inis.run/sessions/{session_id}/sse

Same Bearer auth. session_id is implicit — tools operate on that session. The API returns mcp_url as the base (…/sessions/{id}); append /sse for the client connect URL.

Session tools (20): execute_code, run_command, read_file, write_file, list_files, write_files, find_files, grep_files, mkdir, remove, rename, expose_port, unexpose_port, get_egress, set_egress, get_session, list_connections, revoke_connection, create_checkpoint, list_checkpoints.

No destroy/pause/resume/fork/artifacts/templates on the session server. run_command / files / expose auto-resume a paused session via the normal API wake path. Note: execute_code on both org and session MCP is the stateless-execution code runner — it does not exec inside the bound session; use run_command for that.

See Custom templates & MCP for the create-template → session MCP flow.

When to use MCP vs SDK

Use remote MCP when…	Use SDK when…
Your runtime supports remote MCP (Cursor, Claude Code, AI SDK)	You need fine-grained session control in app code
You want zero integration code and no local install	You are building a custom agent loop
Your agent runs in the cloud, not on the user's laptop	You need custom retry / billing hooks

We do not ship a local stdio MCP server. Use the remote MCP endpoint above, or the SDK.