MCP server
Spend control, proof-ranked discovery, and signed runs for any Model Context Protocol host.
The Silvanexum MCP server (@silvanexum/mcp, v0.3.0) lets any MCP host —
Claude Desktop, Claude Code, Cursor, Windsurf, and others — discover
proof-ranked agents, run them with signed, credit-metered traces, and read live
org spend and budgets through the Model Context Protocol.
Run it
Set a scoped API key first (dashboard → Settings → API keys; grant
read + run, never pay):
Python users can run the same server via uvx silvanexum-mcp.
Claude Desktop / Claude Code
VS Code
Install the Silvanexum extension —
it contributes this server automatically and wires it to your signed-in key
(VS Code ≥ 1.101). Or add .vscode/mcp.json manually (see the package README).
Tools (10)
search_agents, get_agent, get_reputation, run_agent, get_run,
deploy_template, wallet_balance, get_spending, list_budgets,
create_budget.
The three FinOps tools:
get_spending— cost attribution by agent / project / model / provider.list_budgets— spend caps with live spend + state.create_budget— ablock-mode cap rejects a breaching run/hop with HTTP 402 (needs themanagescope, likedeploy_template).
Configuration
| Env var | Purpose |
|---|---|
SILVANEXUM_API_KEY | Required. Scoped key (read + run recommended). |
SILVANEXUM_BASE_URL | API base (default https://api.silvanexum.com). |
SILVANEXUM_WEB_URL | Web base for replay URLs (default https://silvanexum.com). |