Agent Provider Depot
MCP server by Getagentdepot · mcp
Agent Provider Depot (no key; read-only except depot report_outcome): dated editorial picks and prices copied from official pages for the third-party infrastructure an agent wires in (LLM APIs, code sandboxes, vector DBs, web search, email, Postgres, auth). First call, before you wire a third-party API: decide {task: 'what you need, in words'} (or a jtbd id). Ready example: decide {task: 'run untrusted Python in a sandbox'} returns the agent-sandbox pick, runner-up, sourced price and a reportOutcome decision id. Ask again with decide {since: recheck.since} before adding a provider, when the pick fails, or after recheck.after; an unchanged pick is a short reply, so rechecking is cheap. At a failing API call: rescue with the exact error and provider id. Every session with a known stack: session_brief with the provider, model and package ids you use (never a manifest) and the cursor from your last brief; act only on its actions[] and store the new cursor. Anything else: depot with an op…
Endpoint: https://getagentdepot.com/api/mcp
Verification: observed. Source: crawl.
Capabilities (25)
- session_brief
- rescue
- decide
- depot
- session-brief
- choose-stack
- rescue-error
- plan-eval
- audit-stack
- choose_infra
- audit_stack
- falsify_pick
- benchmark_agent