[ Context Review ]

Save tokens. Check what survives.

Select a smaller context pack and check the evidence you say it must retain.

This deterministic review does not call an AI model, charge a wallet, or prove answer correctness. Use the local workbench for optional analysis with your ChatGPT plan; commercial website sign-in is not enabled here.

Question or task

Sanitized documents (JSON array)

Required evidence (exact excerpts, JSON array)

You define what must survive. An empty list does not test completeness. Excerpts are case-sensitive and checked within selected passages from the named source.

Estimated token budget

I confirm these excerpts are sanitized, authorized or synthetic. Do not submit credentials, government IDs, payment card details, protected health information or confidential personal records. This sends the supplied text to Maha’s server for transient computation.

Use from an agent

The Context Review MCP endpoint accepts only explicitly supplied sanitized text. It does not access company accounts, fetch URLs, or sell services.

https://www.mahastrategies.com/api/mcp/context-review

Add the HTTPS endpoint as a remote Streamable HTTP MCP connector in a compatible client. No Maha account or API key is required.
Discover and call review_context_retention. Supply a task, documents with stable IDs and text, a token budget between 64 and 16,000, and sanitized: true only after removing restricted data.
Declare expectedEvidence as source IDs paired with exact, case-sensitive excerpts. Inspect each retention status; an empty list means completeness was not tested.
Review the selected context and limitations. The tool does not prove document authority, semantic completeness, downstream prompt delivery or answer correctness.

For a synthetic first test, use the original SLA and amendment shown above, ask which rebate applies, and require excerpt “25%” from source “amendment-3”. The retained excerpt is a text-selection result, not a legal conclusion. Input is limited to 64 KiB and eight sources.

Data handling · Usage boundaries · Support