The MCP connector lets you work with your Junior calls inside Claude or ChatGPT. MCP stands for Model Context Protocol: it lets an AI assistant connect to another tool and use its data in a conversation, without you copying it across.
Once connected, your assistant can retrieve call details, summaries, and transcripts from the project and personal folders you have access to. If you've used JuniorGPT, the workflow will feel familiar, with a few limitations.
Not connected yet? Start with Set up the connector, or send that page to your IT team.
Give the assistant enough detail
Start with the project and the expert you're asking about. Similar names and topics can appear across projects, so a little context helps the assistant find the right call.
- Project name: if you don't know the exact name, ask the assistant to list your projects first. Call lists default to calls you created, which may leave out colleagues' calls. To include them, say "across the whole team". You still only see calls you have access to.
- Expert's full name: use their first and last name to avoid matching someone else.
- Other call details: the date, job title, organization, country, segment, source, or rating can narrow the search. Use details already recorded in the Call Tracker.
Instead of "What did John say about pricing?", try:
What did John Alderman, the CFO at Meridian Foods we spoke to via GLG, say about pricing on the Meridian project?
Find the right calls
Ask the assistant to filter calls by client or organization, job title, country, expert-network source, segment, rating, cost, duration, or date range. You can combine filters.
Show me all done calls on Project X with CFOs or Finance Directors, sourced from GLG, since 1 May.
Which calls in Project Y are rated 4 or 5?
Pull a summary or transcript
Ask for Junior's Key Takeaways and Call Summary by name when you want the versions already in Junior.
If you ask for a summary without specifying Junior's Key Takeaways or Call Summary, the assistant may write its own. That can introduce errors, so be clear whether you want to retrieve an existing summary or create a new one.
Pull the Junior Key Takeaways and Call Summary for the call with [expert name / organization] on [date].
Find one call first, then ask about a topic or quote from it:
In that call, what did the expert say specifically about switching costs? I need the exact wording.
The connector doesn't support keyword or semantic search across transcript text. The assistant reads the selected call's full transcript to find the passage. It can also retrieve a project's calls together, but several full transcripts can exceed what it can handle. Narrow the selection or ask for summaries and Key Takeaways first.
Compare what experts said
You can ask the assistant to identify themes, compare views, and flag evidence that contradicts your working hypothesis. Keep the scope manageable and check the source calls before using the answer.
Across the done calls on the market-entry workstream, what is the range of views on regulatory risk? Flag any experts who disagree with the majority.
Summarize what industry buyers versus competitors said differently about pricing power in the [segment] segment.
Prompts to try
These prompts also work in JuniorGPT.
| When you need to… | Try asking… |
|---|---|
| Prepare for a call | Before I speak to [expert], what have other experts on this project said about [topic], so I don't repeat ground we've covered? |
| Test a hypothesis | Do the calls so far support or contradict [hypothesis]? Show me the calls on each side. |
| Check coverage | Which job titles, geographies, or segments have we not spoken to yet on this project, based on our interviews so far? |
| Research a competitor | What have competitors' former employees said about [competitor]'s go-to-market model across our projects this year? |
| Find a quote for a deck | Find one strong, verbatim quote on [topic] from a rated-5 call, with the expert's title and organization for attribution. |
| Pick up a workstream | Give me a rundown of the calls conducted so far and the headline takeaway from each. |
What stays in Junior
The connector is useful for working with calls in another assistant, but it doesn't replace Junior's research tools.
- Answers grounded in your project: Claude or ChatGPT may mix call content with web results or other documents, increasing the risk of an incorrect answer. JuniorGPT draws on the calls and files in your project.
- Reliable inline citations: JuniorGPT links answers to the call notes and the exact passage behind a quote or insight. Claude and ChatGPT don't provide the same reliable path back to the source.
- Aggregated insights: the Insights Grid, Running Summaries, and Saved Quotes stay in Junior. The connector can read a call's Key Takeaways and identify entities marked as competitors, but it doesn't expose those continuously updated project views.
See the MCP tool reference for everything the assistant can retrieve.