Argosvix — AI agent observability (87 tools)

What is Argosvix — AI agent observability (87 tools)?

Argosvix — AI agent observability (87 tools) is an MCP server that lets AI agents (Claude Desktop, Cursor, Codex CLI, custom MCP clients) query, manage, and operate their LLM observability data directly from the conversation. It surfaces 87 tools, 3 resources, 8 resource templates, and 3 prompts, covering read/write operations on call records, cost, alerts, anomaly detection, and runtime control plane features like budget gates and human-approval gates.

How to use Argosvix — AI agent observability (87 tools)?

Install globally with npm install -g @argosvix/mcp-server. Configure your MCP client (e.g., Claude Desktop or Cursor) by setting the ARGOSVIX_API_KEY environment variable in the server config. Optionally, start with --http for remote HTTP transport on port 3000. The server supports both stdio (subprocess) and HTTP (remote/self-host) transports.

Key features of Argosvix — AI agent observability (87 tools)

• 87 tools covering observability read/write and autonomous AI-ops actions (e.g., get_account_health, detect_anomaly, propose_alert_rules)
• Runtime control plane with budget gates, policy gates, and human-approval gates
• HTTP transport with per-request API key auth and security notes
• Profile toggling (ARGOSVIX_MCP_PROFILE=core) to expose only 11 essential tools for tight context budgets
• Resource subscriptions for real-time updates (stdio only, poll every 60s)
• English and Japanese descriptions via ARGOSVIX_MCP_LANG environment variable

Use cases of Argosvix — AI agent observability (87 tools)

• Query recent LLM call cost, tokens, or latency by provider or model directly in the agent conversation
• List, create, silence, or acknowledge alerts without switching to a dashboard
• Audit cost trends or investigate error/latency anomalies using the provided prompts
• Manage budget gates, policy gates, and alert rules autonomously from the AI agent

FAQ from Argosvix — AI agent observability (87 tools)

What exactly can the agent do?

The server exposes 87 tools (84 generally available + 3 founder-ops) for reading and writing observability data—call records, costs, alerts, anomaly detection, and runtime control plane actions. It also provides 3 resource templates for dynamic lookups and 3 prompt templates for common workflows.

Do I need an API key?

Yes, for full operation you need a valid Argosvix API key (argk_...). Without it, the server still starts in introspection-only mode: all 87 tools are listed but any tool call returns instructions to get a key at https://dashboard.argosvix.com/api-keys.

What transports are supported?

stdio (subprocess) and HTTP (remote/self-host). stdio supports resource subscriptions for live updates; HTTP is stateless with per-request API key auth.

Is there a way to limit the number of tools exposed?

Yes. Set ARGOSVIX_MCP_PROFILE=core to expose only 11 essential tools for day-to-day operations (e.g., query_calls, get_cost_summary, list_alerts). The default profile is full (87 tools).

How is privacy handled?

The MCP server sends queries to Argosvix’s ingest endpoint using your API key. No prompts or completions are exposed—only metadata (tokens, cost, latency, model name, your tags).