Loading…
Loading documentation...
Wire Claude Code, Codex, Cursor or Gemini to /api/mcp/v1 with 129 tools, scoped tokens, resource locks and AI write guardrails.
Commands checked on 2026-11-12 with MCP 2025-11-25 Streamable HTTP, @serversareup/mcp-server 0.2.0.
Your AI coding agent or desktop assistant—Claude Code, OpenAI Codex, Cursor, Windsurf, Gemini CLI, or custom MCP client—will connect to Servers Are Up's Hosted Streamable HTTP MCP Server (https://serversareup.com/api/mcp/v1, exposing 129 tools) or local stdio package (@serversareup/mcp-server, 35 tools), protected by least-privilege token scopes, CanNotDelete/ReadOnly Management Locks, and server-side AI Agent Write Guardrails.
claude), OpenAI Codex (codex), Gemini CLI (gemini), Claude Desktop, or Cursor.5 public tools: sau_search_catalog, sau_recommend_plan, sau_compare_providers, sau_search_docs_and_tools, sau_get_platform_facts).124 authenticated tools), mint a scoped sau_... token at Account settings > API keys (/account/api-keys).Verify that your network reaches the Hosted Streamable HTTP MCP endpoint (https://serversareup.com/api/mcp/v1) with a standard JSON-RPC 2.0 call—no Authorization header required for public catalog, provider comparisons, and documentation search:
curl -fsSL -X POST "https://serversareup.com/api/mcp/v1" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "sau_recommend_plan",
"arguments": { "workload": "nextjs postgres vector rag", "budgetUsd": 45 }
}
}' | jq .
Before handing a token to an autonomous agent:
/account/api-keys), start with read-only scopes (instances:read, network:read, storage:read, cloud:read, activity:read, account:read, catalog:read). Grant instances:exec, instances:write, or cloud:write only when you want the agent to run commands or provision cloud resources./dashboard/resource-manager), switch to the Portal & AI Settings tab, and choose your AI Agent Write Guardrail mode:READ_ONLY: Blocks all mutating MCP/AI write operations across Resource Manager and Cloud planes even if a token carries :write scopes.SAFE_MUTATIONS: Permits additive writes (creating tags, groups, or configuration keys) while blocking destructive deletes.FULL_AUTONOMOUS: Allows full scoped automation (while still strictly enforcing CanNotDelete and ReadOnly locks!)./api/mcp/v1 in Claude Code, Codex, Gemini, or CursorAdd the hosted HTTP MCP server with a single command in your preferred agent CLI:
# Claude Code (Streamable HTTP + Bearer token)
claude mcp add --transport http serversareup https://serversareup.com/api/mcp/v1 \
--header "Authorization: Bearer sau_your_token_here"
# OpenAI Codex CLI
codex mcp add serversareup --url https://serversareup.com/api/mcp/v1
# Google Gemini CLI
gemini mcp add --transport http serversareup https://serversareup.com/api/mcp/v1 \
--header "Authorization: Bearer sau_your_token_here"
For Cursor (.cursor/mcp.json) or Claude Desktop, configure the serversareup server entry—or install the official Claude Code Plugin (https://serversareup.com/.claude-plugin/marketplace.json), which bundles the serversareup skill and automatic read-only tool approval hooks (packages/claude-plugin).
If your MCP host only speaks local stdio, download and run @serversareup/mcp-server (https://serversareup.com/downloads/serversareup-mcp-server.tgz) with SAU_API_TOKEN=sau_your_token_here.
Once connected, your agent has native schema-typed tools across every plane of the platform—including sau_cli_execute (which runs any structured sau CLI command over JSON), sau_rm_* (Resource Groups, Tags, Locks, RBAC, Diagnostics, Quotas, HCL/JSON export), sau_cosmos_*, sau_blob_*, sau_vault_*, sau_sentinel_*, sau_exec_instance, and sau_list_instances.
Ask your agent concrete, verifiable prompts such as:
staging-ai-rg, deploy a VECTOR_AI Cloud Database named kb-vectors, and export the group as Terraform HCL."Run claude mcp list (or ask your agent "Call sau_get_platform_facts and sau_list_instances"), then verify the audit trail on Operations > Activity log (/dashboard/activity) or via the CLI:
sau monitor activity-log --limit 10 -o table
Every MCP tool call is logged with the caller token, timestamp, resource URI, and result.
401 Unauthorized when calling authenticated tools (sau_list_instances). Zero-auth tools work without headers, whereas account tools require Authorization: Bearer sau_.... Confirm there is no typo in the header name or token string.403 AI_WRITE_GUARDRAIL_BLOCKED on a mutation. Your Portal & AI Settings mode is set to READ_ONLY or SAFE_MUTATIONS on /dashboard/resource-manager. Keep that guardrail enabled for exploratory sessions, or switch it temporarily when running planned provisioning.409 RESOURCE_LOCKED. A CanNotDelete or ReadOnly lock prevented the agent from deleting or altering a protected production resource—working exactly as designed!CanNotDelete locks (sau lock create ... --level CanNotDelete) to every production server, database, and storage container before giving any AI agent :write or :exec scopes.AI Automation Operator) on /dashboard/resource-manager scoped only to the specific Resource Group (staging-ai-rg) the agent owns.https://serversareup.com/api/mcp/v1) over local builds so your agent automatically receives all 129 tools and zero-downtime schema updates.Every Servers Are Up virtual machine and container includes an integrated HTML5 browser-based SSH, RDP, and VNC console in your dashboard, default-off inbound firewall rules with per-IP allow-lists, and full programmatic control via REST API v1 and Model Context Protocol (MCP). Explore our Virtual Datacenter Plans or reach out to Servers Are Up Support.