Generic assistants write confident nonsense about evaluation criteria. Writers stop trusting them, go back to the template, and the assistant becomes an expensive spell-checker.
Ground every answer in retrieved guidance rather than model memory: the knowledge base returns curated programme and evaluation material.
Pin the call itself into context so impact and excellence sections argue against real expected outcomes.
Force citations. A sentence the writer cannot trace back to a source is a sentence they will not paste.
- 01search_knowledge_base
Retrieve programme, evaluation and consortium guidance for the section being written.
- 02get_fund_summary
Anchor the draft to the actual expected outcomes and conditions of the call.
- 03search_cordis_projects
Cite precedent when the section needs evidence of feasibility.
“Draft the impact section for this call. Use the official expected outcomes, and cite two funded projects that show the approach works.”
curl -X POST https://enabler.heeyuu.eu/api/mcp/protocol \
-H "Authorization: Bearer YOUR_HEEYUU_KEY" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_knowledge_base","arguments":{"query":"how is impact assessed in Horizon Europe evaluation","limit":3}}}'