Use these questions to decide whether Toolcaise fits your operation. Each answer says what works today and exactly where it stops.
Available today
Scoped credentials over outbound HTTPS: Python and Node.js SDKs, the HTTP contract, OpenTelemetry traces, or Toolcaise Connect on the machine that runs the agent.
Where it stops
Each path needs a setup step. Toolcaise does not discover agents on its own.
Available today
Run history, step spans, tools, retries, timing, and token usage and costs.
Where it stops
Through the SDKs and OpenTelemetry these are reported by the agent or framework. A missing trace is not proof that no work happened.
Available today
Incidents for stale heartbeats, overdue schedules, stalled runs and repeated failures, with alerts to the Slack, webhook or email destinations your workspace sets up.
Where it stops
Detection depends on reports arriving. Closing an incident does not mean its condition cleared.
Available today
Reported outcomes with a separate human acceptance or rejection and a review history.
Where it stops
Acceptance records a decision. Inspect the actual output or receipt separately.
Available today
Toolcaise Connect enforces cost, token, model-call and runtime limits on a supervised run and stops the whole process tree at the limit, with or without a connection to Toolcaise.
Where it stops
Token and cost limits cover OpenAI- and Anthropic-compatible traffic through the standard base-URL settings, and cost uses prices you supply. Figures reported through the SDKs are visibility, not a cap.
Available today
A Connect supervised run started with --remote-control can be paused, resumed and stopped from the dashboard. SDK runners support cooperative pause, resume and cancel.
Where it stops
SDK controls apply only at checkpoints your code implements. Connect picks up a stop on its next poll: within about ten seconds, or up to two minutes after failed polls.
Available today
Guarded actions bind approval to a run and an action digest, consumed once before execution. The MCP tool gateway allows or denies tools by name, and can hold a tool call for approval bound to its exact arguments.
Where it stops
Only guarded operations and MCP tools set to require approval participate. The approver of an MCP call sees the tool name, not its arguments.
Pick the path by what you need to enforce, not only by what you want to see. You can use more than one.
| Path | What Toolcaise sees | What it can enforce |
|---|---|---|
| Python and Node.js SDKs, HTTP | What the agent reports: runs, steps, usage, outcomes | Pause, resume, cancel and exact-action approvals, at checkpoints your code implements. Cooperative: code that skips the checkpoint is not stopped. |
| OpenTelemetry | Runs, model and tool spans, and token counts from your framework’s own tracing | Nothing. It observes only. |
| Connect: Claude Code and Codex adapters | The allow-listed events and metrics those tools emit | Nothing. It reports telemetry only. |
| Connect: supervised run | Every model call made through an OpenAI- or Anthropic-compatible API, measured on your machine, plus the process’s start, end and resources | Runtime, cost, token and model-call limits and a model allowlist, by stopping the whole process tree. Pause, resume, stop and limits from the dashboard. |
| Connect: MCP tool gateway | Every tool call that passes through the proxy | Allow or deny by tool name, a call budget per session, and a hold until a person approves the exact call (Connect 0.3 or later). The approver sees the tool name; the arguments stay on your machine. |
Connect the agent, report its next run, and inspect where the work changed or failed. Then add the review, limit or checkpoint your team actually needs. Time to diagnosis and reviewed outcomes are more useful pilot measures than a feature count.
Create a free workspace, connect the agent you already run, and watch its next run arrive. Free covers 3 agents, with no card required.