Product Updates
DiscoLike MCP: Refine Your ICP Query Plan
DiscoLike MCP exposes the query plan behind an ICP search so an agent can review and revise the search text, phrase matches, and filters before expanding the results.
· updated October 6, 2026
A target account list is easier to improve when you can see the query that produced it. DiscoLike MCP returns the search plan behind an ICP request so the assistant can inspect its assumptions and revise them.
That makes the conversation useful beyond the first request. You can explain why a result is wrong and adjust the corresponding part of the search.
What is in the query plan?
The plan can include semantic search text, phrase matches, industry groups, geography, and technology constraints. These represent different parts of the ICP.
For example, “managed IT providers serving law firms” describes business fit. A country is a hard constraint. A named certification may require a phrase match or additional verification.
Keeping the parts visible helps avoid burying every requirement in one paragraph.
How the feedback loop works
- Send the assistant a description of the target customer and relevant context.
- Request a proposed DiscoLike search plan.
- Review the assumptions and inspect sample results.
- Revise the text or filters that caused mismatches.
- Run the larger search after the sample meets your criteria.
Give the agent explicit acceptance criteria and keep a record of the query you approved. If repeated revisions do not improve the sample, revisit the business definition or source coverage.
What the Outcome Learning Model contributes
DiscoLike’s query assistant uses patterns from prior customer searches to propose a starting configuration. Those patterns help translate a rough business description into search parameters.
Your client context still matters. A historical pattern may be a poor fit for an unusual business model or a new campaign requirement.
Use feedback to diagnose the right problem
If relevant companies are absent, inspect the source coverage and restrictive filters. If the list contains adjacent business types, refine the semantic description. If evidence is incomplete, queue validation rather than forcing a yes-or-no decision.
Explore DiscoLike MCP to put query review inside your agent workflow.
Related posts:
