A saved filter becomes useful when the same research question returns regularly. It also creates a quiet risk: a one-off experiment can overwrite the version you intended to keep. Next week's result then looks comparable even though the method changed.
Separate three actions in your research notes: reading a filter, running it and changing its saved definition. OddsTips MCP exposes these as different operations. The following workflow is based on its documented personal-filter behaviour.
Read the definition, not just its name
A label such as “Home favourite” does not tell you whether the filter uses opening prices, closing prices or a reference-match tolerance. Ask the assistant to retrieve the owned filter and explain each criterion before analysing a fixture. Check the sport, source and period as well as the numerical bounds.
For a fixed range, keep both endpoints. For a reference-based comparison, record which fixture supplied the reference values. Reusing a filter does not necessarily freeze the original fixture's prices: an ordinary comparison can be a template that resolves against the next reference match.
Try a temporary variant
Read my selected filter. For this run only, use a fixed home-win opening-odds range of 1.90 to 2.20. Keep the other applicable criteria. Explain any incompatibility and do not save the variant.
This example is illustrative. It does not assert that a filter with those fields already exists. The assistant should inspect the actual columns and operators rather than inventing IDs or deleting a criterion that it cannot apply.
Compare the original and variant with the same reference context and date scope. Record which criterion changed. If the variant returns more rows, that is evidence of a different selection boundary, not automatically a better method.
Save only after naming the change
If you decide to keep a variant, state whether you want a new private filter or an edit to the existing one. A new name can make the distinction clearer: “Opening range 1.90–2.20, review A” carries more context than “Better filter”. Avoid putting tokens or other private credentials in filter names or descriptions.
The documented update process uses the latest filter revision. If another device has edited it meanwhile, the assistant should read the newer version and explain the conflict. Silently overwriting it would discard a decision made elsewhere.
Do not turn a connection problem into a second analysis
After a timeout, first establish whether the original analysis was accepted. An accepted job should be checked using its existing result identifier. Retrying an ambiguous operation as a completely new request can repeat work and consume another allowance. This is different from deliberately asking for a second comparison.
Keep a compact comparison record
Write down the original criteria, changed criterion, reference fixture, retrieved sample counts and missing-data notes. If the same source and period later return a different count, those details help distinguish a filter edit from an archive update. They also make clear which findings were exploratory and which method was kept for future use.
Sources and method
Original research examples and prompts, not observed match results or performance claims. Prepared with AI assistance; source statements, examples and calculations were checked. Product details were checked against the OddsTips MCP setup and usage guide, version 0.3 dated 12 October 2026, and the corresponding product source. Client support and account access must be checked at use time. The product link opens the homepage; choose MCP in its navigation.
- OddsTips homepage — open MCP in the navigation for the product guide — accessed 13 October 2026