> ## Documentation Index
> Fetch the complete documentation index at: https://docs.iqrush.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Manage compatibility and interface differences

> Keep a small set of contract checks in your integration so an added field or a changed deployment does not silently change a report.

Keep a small set of contract checks in your integration so an added field or a changed deployment does not silently change a report.

## Discover the current contract

Refresh the MCP tool catalog when connecting or updating a client. Validate outgoing names and types against `inputSchema`. This documentation edition covers 43 tools from the reviewed Dev source. Deployment catalogs can differ, so a count is not a capability guarantee.

Treat project IDs and run IDs as strings, tolerate added response fields, and preserve nulls. Validate units and response scope before formatting values. Do not infer metric meaning from a field name shared by another API.

`schema_version: "1.0"` is the measurement envelope's version label. It is separate from the negotiated MCP protocol version, the host application's version, and an IQRush release number.

## Known differences in this edition

| Area | Behavior to account for |
| - | - |
| Mention defaults | MCP defaults to coverage/entity; HTTP defaults to count/bundle. Set both explicitly. |
| Sentiment defaults | MCP defaults to positive/entity; HTTP defaults to net/bundle. Set both explicitly. |
| Source comparison size | MCP defaults to 10 sources per brand; HTTP defaults to 25. Set `top_per_brand` explicitly. |
| Mention coverage at bundle level | Read the served aggregation. It can be entity-level even when bundle was requested. |
| Sentiment aggregation | Rows remain entity-level when bundle is requested. |
| Citation URL grouping | URL-level citation scores do not apply group\_by in this edition. |
| Grounding-source denominator | per\_source is accepted, but the reviewed handler changes the label without calculating a different rate. Use grounding\_pool for the documented workflow. |
| Grounding-source subject | The HTTP subject parameter is not applied. Use brand\_subset for supported brand selection. |
| Grounding-source grouping | Platform grouping is supported for all brands with unrestricted sentiment; narrower selections can drop it with a warning. |
| Time-series filters | HTTP supports `filters`; the MCP tools in this edition do not expose that argument. |
| Search Console device | The MCP `device` argument is not applied by the current HTTP route. |
| Technical feature domain | The MCP `feature_domain` argument is not applied by the current HTTP route. |
| Technical recommendations | Available on the documented HTTP read route; no matching external MCP tool in this catalog. |
| Baselines | `list_baselines` returns an empty list; `analytics_mode.baseline_id` is unsupported. Use explicit comparison runs. |

## Check a release with representative cases

Use a project you are authorized to read. Confirm project discovery, a ready run, one measurement with a confidence interval, one grouped result, an empty or unavailable result, and any comparison your application relies on. Compare matching run IDs and scopes before comparing the UI with a connector result.

If two interfaces disagree, preserve both responses and their exact scopes for investigation. Do not quietly average them or choose whichever result better supports the report.

Next: [Availability and limits](/developers/guides/22-availability).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.