---
title: "Start"
description: "Session orientation tool: surfaces tier, capability summary, 6-item action menu, and a structured question before any other tool runs."
icon: material/rocket-launch
---

The first tool the agent calls whenever a user expresses a cost-cutting goal or open-ended platform question in a fresh session. Probes the environment (Reporter, Receiver, Retriever, log analyzer) and returns a full orientation envelope:

- **tier** detected (dev / reporter / receiver / retriever)
- **capability_summary** (what is and isn't available without further setup)
- **action_menu** with 6 items, each gated by the detected tier
- **must_render_verbatim**: pre-rendered markdown the agent surfaces as-is
- **must_ask_user**: the numbered question the user must answer before any follow-up tool runs
- **forbidden_next_actions**: tools the agent must not call until the user picks

## :material-code-braces: Example

!!! tenx-ask "You"

    where do I start?

!!! tenx-answer "Log10x"

    ### Log10x orientation

    **Tier:** Reporter, cost attribution metrics flowing; in-path actions not yet wired.

    **Journey:**
      [x] 1. Visibility
      [ ] 2. Attribution
      [ ] 3. Mitigation
      [ ] 4. Overflow
      [ ] 5. Commitment

    **What would you like to do?** Pick a number:
      1. Show me what cutting costs would look like (estimate before action)
      2. Investigate a specific cost spike or top-N pattern right now
      3. Fetch back events from your overflow bucket (incident / audit / debugging)
      4. Deploy the Receiver so 10x can compact, offload, or tier down in-flight (keep everything), or sample/drop
      5. Set up the overflow bucket (your own S3); diverts noisy patterns out of the SIEM for cost savings, events stay recoverable
      6. Just orient me, explain what 10x does and what I should ask next

    _(Pick a number before I run any other tool.)_

## :material-chat-question-outline: More to ask

- *"orient me, first session"*
- *"I want to cut my Splunk bill"*
- *"what can 10x do?"*

## :material-check-decagram-outline: Prerequisites

None. The tool degrades gracefully when no environment is configured.

## :material-code-json: Schema and samples

??? tenx-input-schema "Input schema"

    Agent-facing JSON Schema (the canonical shape the MCP server publishes via `tools/list`):

    ```json
    {
      "type": "object",
      "properties": {
        "intent_hint": {
          "type": "string",
          "enum": [
            "cost",
            "forensic",
            "install",
            "orient"
          ],
          "description": "Optional hint at what brought the user here. `cost` = wants to cut/save bill; `forensic` = wants to read back the offloaded/overflow cohort from the customer-owned S3 bucket; `install` = wants to deploy; `orient` = first-time session, open-ended. When omitted, the tool treats it as `orient`."
        },
        "session_state": {
          "type": "string",
          "enum": [
            "fresh",
            "midway",
            "returning"
          ],
          "description": "Optional caller hint at where the user is in the session. `fresh` (default when omitted) = first message of a new chat, emit the full orientation envelope. `midway` = the agent has already surfaced the orientation envelope this session and the user is mid-flow, emit a short \"already oriented, proceeding\" envelope. `returning` = treated as `fresh` because the menu / journey phase may have shifted since the last session."
        }
      },
      "additionalProperties": false
    }
    ```

    Source: [`src/tools/log10x-start.ts`](https://github.com/log-10x/log10x-mcp/blob/main/src/tools/log10x-start.ts).

??? tenx-output-schema "Output schema"

    The `data` block inside the [StructuredOutput envelope](../index.md#json-by-default-output):

    ```typescript
    interface ToolData {
      tier: 'dev' | 'reporter' | 'receiver' | 'retriever';
      siem_detected: string | null;
      capability_summary: {
        cost_attribution_available: boolean;
        compact_installable: boolean;
        tier_down_available: boolean;
        forensic_query_available: boolean;
        offload_ready: boolean;
        siem_query_available: boolean;
        receiver_discrimination_uncertain: boolean;
      };
      action_menu: Array<{
        action: string;
        label: string;
        applicable: boolean;
        gated_reason?: string;
        routes_to: string;
      }>;
      journey_phases: Array<{
        phase: number;
        name: string;
        description: string;
        current_status: 'complete' | 'in_progress' | 'not_started' | 'blocked_by_prior_phase';
      }>;
      must_render_verbatim: string;
      must_ask_user: { question: string; options: string[] };
      forbidden_next_actions: string[];
      intent_hint: 'cost' | 'forensic' | 'install' | 'orient';
    }
    ```

    Envelope-level fields the agent should also read: `summary.headline` (1-line answer), `actions[]` (next-call chain hints as `{tool, args, reason}`), `schema_epoch` (engine-ID stability boundary).