Medeloop Grants

Confirm an onboarding file upload by updating DynamoDB references.

Record a completed upload in the database.

Two-step flow: (1) get_upload_url returns a presigned S3 PUT URL, (2) after the widget uploads the file to S3, this tool records the upload metadata (key, filename, size) in the ProposalDocument table.

Create a new proposal and return its server-generated ID.

Delete an onboarding file from S3 and DynamoDB.

Export proposal to Word (DOCX) format.

Export proposal to PDF format (via DOCX → LibreOffice conversion).

Find Custom Grants

Opens an interactive grants discovery widget that finds grants personalized to the researcher's profile.

When to Use

Use this tool whenever the user wants to:

• Find grants ("find me grants", "search for grants")
• Get grant recommendations ("what grants match my profile?")
• Search for specific grants ("find NIH grants about cancer")

How It Works

This tool opens a widget that: 1. Checks if the user has completed their researcher profile setup 2. If not, guides them through onboarding first 3. Then searches for grants using their profile + any specific queries you provide

Parameters

• queries (array of strings): Search terms for grants
• Empty array [] = search based on user's profile/research interests
• Specific queries = search for those terms, enhanced with profile context

Examples

Profile-based search (user says "find me grants")

find_custom_grants(queries=[])

Specific search (user says "find cancer research grants")

find_custom_grants(queries=["cancer research", "oncology", "tumor biology"])

Combined search (user says "find NIH grants about machine learning")

find_custom_grants(queries=["machine learning", "artificial intelligence", "deep learning"])

Important Notes

• Always use this tool for grant discovery - it handles onboarding automatically
• The widget will execute the actual search after ensuring the user is onboarded
• Do NOT use search_grants directly - that's for internal widget use only
• After results are shown, the user can interact with the widget to explore grants

Generate AI-driven follow-up intake questions based on Round 1 answers.

Generate a grant template based on opportunity number.

Retrieve intake state: Round 1/2 answers and revision.

Get my biosketch as a string

Get my CV as a string

Get my previously written grants relevant to the query as a list of {file_name, text}

Get my research interests and topics. Use this tool when user asks me to find recommended grants for my profile

Get my key recent publications relevant to the query as a list of {file_name, text}. These are key publications that I uploaded to the system.

Get my researcher information (name, position, institution, research interests)

Get my publications relevant to the query as a list of {file_name, text}. These are publications found from my Google Scholar profile.

Get the current user's onboarding status and profile data. Returns onboarding completion status and all profile fields.

Field names match medeloop-app-api DTO format (camelCase).

Get a presigned S3 PUT URL for onboarding file upload.

Widget flow: 1. callTool("get_onboarding_upload_url", {...}) → {url, key} 2. fetch(PUT, url, file_body) — direct to S3 3. callTool("confirm_onboarding_upload", {...}) — if needed

Return proposal restore status for refresh handling.

Generate a presigned S3 PUT URL so the widget can upload directly.

List the current user's proposals.

Poll for new proposal section updates from the database.

Retrieve writing instructions for a specific proposal section.

Section can be one of:

• 'specific_aims'
• 'significance'
• 'approach'
• 'others'

Save user intake answers for a proposal.

Persist a user's manual edit to a proposal section.

Search Grants

Search for research grants using semantic search with multiple queries.

This tool accepts an ARRAY of search queries and automatically executes all queries, combines the results, and deduplicates them for you.

How to use this tool

FOR SPECIFIC SEARCH QUERIES

• Provide an array of targeted keywords related to the research area
• Search is semantic - understands context and intent, not just exact keywords
• Example: search_grants(queries=["machine learning computer vision", "deep learning neural networks"])

FOR PROFILE-BASED SEARCH

When user asks "find grants" or "grants for my profile":

**MANDATORY WORKFLOW

• You MUST follow ALL these steps:**

1. **Gather researcher context

• THIS IS CRITICAL:**

Call get_my_research_context() to get complete researcher profile:

• Basic info (name, title, institution)
• Research interests and topics
• Biosketch (qualifications and experience)
• CV (complete background)
• Publications (recent work to understand expertise)
• Previously written grants (for context)

This single tool call replaces the need to call 6 separate tools and provides all the context needed to generate targeted grant queries.

2. Analyze and identify themes:

• Extract primary research areas from interests and publications
• Identify methodologies (e.g., "computational modeling", "field studies", "lab experiments")
• Note key themes and technical approaches from publications

3. Generate 2-3 DIFFERENT targeted queries:

• Query 1: Primary research area + methodology
• Query 2: Specific domain + technical approach
• Query 3: Alternative angle or subdomain from their work
• Make each query distinct to capture different relevant opportunities

4. **Execute search with multiple queries

• THIS TOOL HANDLES DEDUPLICATION:**
• Call search_grants() ONCE with an array of 2-3 DIFFERENT queries
• The tool will execute all searches, combine results, and deduplicate automatically
• Use num_results=20 to get comprehensive results per query
• Example: search_grants(queries=["cancer immunotherapy", "tumor microenvironment", "CAR-T cell therapy"], num_results=20)

5. Present results with context:

• Results are already deduplicated by OpportunityID
• Explain why each grant matches their profile
• Reference specific publications or expertise areas
• Prioritize by relevance score and deadline

Example Profile-Based Workflow

User says: "find me grants"

Step 1

• Call get_my_research_context():
• This returns a complete context object with all researcher information
• Includes info, interests, biosketch, CV, publications, and grants
• Example: context = get_my_research_context(publications_limit=5)

Step 2

• Analyze the data gathered:
• Interests: Identify main research topics and domains
• Publications: Understand methodologies and technical approaches used
• Biosketch: Note years of experience and career stage
• Previous grants: Identify funding agencies and award types

Step 3

• Execute ONE search with 2-3 different queries:

search_grants(
    queries=[
        "[primary research area] [methodology]",
        "[domain] [technical approach]",
        "[alternative angle] [subdomain]"
    ],
    num_results=20
)

The tool automatically combines and deduplicates ~40-60 results from the searches.

Important Rules

• This tool requires a queries parameter with an array of 1+ search terms
• For profile-based search: provide 2-3 different queries in the array
• Each query should use different keywords/angles to maximize coverage
• Tool handles combining and deduplicating results automatically

Presenting Results to User

When presenting results to user, ALWAYS show:

• Title (as clickable link using AdditionalInformationURL)
• Agency
• Opportunity Number
• Deadline
• Funding amount (if available)

Proposal drafting rule (IMPORTANT)

If the user wants to draft a proposal for a grant you found:

• Do NOT draft a proposal in chat.
• Do call write_grant_proposal(opportunity_number=...) using the grant’s opportunityNumber from the search results.
• If the widget provides a stable drafting session key, also include draft_key=... so the same proposal can be resumed after refresh.

Start the background grant writing agent for a proposal.

Submit user answers for a specific proposal section.

Update onboarding profile data.

Use this tool when the user requests to write or draft a grant proposal. You must use this tool when the user requests to write a proposal for a grant you found in the search results. This tool will initiate an interactive grant proposal widget to allow the user to interact with the written proposal.

CRITICAL: You MUST use the 'opportunityNumber' field from search_grants results!

• DO use the 'opportunityNumber' field (like RFA-AI-24-012, PA-24-123, etc.)
• DO NOT use the 'id' field (UUID like b34682f7-2ccc-52b0-abba-00cb42ee40ca)
• If a draft_key is available from the widget or a prior tool output, you MUST pass that exact draft_key unchanged.
• DO NOT invent, modify, or omit a known draft_key when resuming the same proposal session.

Example of CORRECT usage:

• opportunityNumber='RFA-AI-24-012' ✓
• opportunityNumber='PA-24-123' ✓
• write_grant_proposal(opportunity_number='PAR-25-043', draft_key='draft-123') ✓

Example of WRONG usage:

• opportunity_number='b34682f7-2ccc-52b0-abba-00cb42ee40ca' ✗ (this is the id field)
• write_grant_proposal(opportunity_number='PAR-25-043') ✗ when resuming and a draft_key is known
• write_grant_proposal(opportunity_number='PAR-25-043', draft_key='new-key') ✗ if the existing draft_key was different

Some examples of user requests that should trigger this tool:

• "Please draft a proposal for the grant with opportunity number RFA-AI-72-841"
• "I would like to write a proposal for the 3rd grant."
• "Can you write the grant with opportunity number PA-24-123?"