MCP elicitation
MCP elicitation works even when your agent doesn't support it
When an MCP tool stops mid-call to ask a question and your agent can't show the form, Meandr puts the question on your dashboard and returns the answer on the same call.
The problem
MCP elicitation: a tool asks a question your client can't answer
MCP elicitation lets a tool pause mid-call to ask a person for one more thing: a confirmation, a choice, a missing field. The agent's client has to show that form. Many clients can't, and the call stalls or fails.
How it works
Meandr holds the call open until a person answers the server's question
There is nothing to set up. When the agent's client can show the form, it appears in the agent's own chat. When it can't, the question comes to your dashboard.
-
01 The question
Your team sees the server's question in the dashboard
A notification in the console names the server, the agent and the tool, and gives the time to answer by.
- Everyone on the team also gets it by email, with an Answer the question button
- The answer is due a few minutes after the call
1NotificationsA server needs input in MainnowEverything MCP (stdio) asked for input during Claude Desktop DS's call to stdio-trigger-elicitation-request: A server is asking a question
Answer it by October 15, 2026 at 11:17 UTC.
Review request -
02 The answer
A person reviews the call, fills in the form and allows it
The request shows which agent called which tool, under which policy. Allow opens the server's form, and Deny turns the question down.
- Anyone with access to the workspace can answer, not only the person who started the agent
- If nobody answers in time, Meandr declines for you and the call ends cleanly
× A server is asking a questionThe tool asked a question before continuing — this is what it wanted to know.
Call- Agent
- Claude Desktop DS
- Server
- Everything MCP (stdio)
- Tool
- stdio-trigger-elicitation-request
- Policy
- Everything MCP (stdio)
- Requested
- Oct 15, 2026, 11:13 AM
- Expires
- Oct 15, 2026, 11:17 AM
CloseDenyAllow -
03 The result
The agent gets the answer on the same call
Meandr passes the answer to the waiting tool, and the tool finishes as if the agent had answered itself.
- The agent doesn't retry or poll: its call stays open while the person answers
- The server receives exactly what was typed into the form
Used meandr: Trigger Elicitation Request Tool
The tool returned:
✓ User provided the requested information! User inputs: - Name: Test - Agreed to terms: false - Favorite Integer: 42 - Favorite Number: 3.14 Raw result: { "action": "accept", "content": { "name": "Test", "check": false, "number": 3.14, "integer": 42 } } -
04 The record
The question, the answer and who gave it stay with the call
Every question a server asks becomes a request in your workspace. Open it later to see what was asked, what came back, when it was decided and by whom.
- In Traffic, the call carries the answer in its response and a link back to the request
- View form shows the answers exactly as they were sent
× A server is asking a questionThe tool asked a question before continuing — this is what it wanted to know.
Call- Agent
- Claude Desktop DS
- Tool
- stdio-trigger-elicitation-request
- Outcome
- Completed · HTTP 200 · 127220 ms
- Requested
- Oct 15, 2026, 11:13 AM
- Expires
Oct 15, 2026, 11:17 AM- Decided
- Oct 15, 2026, 11:15 AM
- By
- DS (ds@meandr.com)
srv Everything MCP (stdio)2026-10-15▾11:15:16200127.22sallowPolicy Everything MCP (stdio)From 203.0.113.24View approval →⌄Response 777 BResult
User provided the requested information! +2 morename: Testinteger: 42number: 3.14resultType: complete
Questions about MCP elicitation
What if my client does support elicitation?
What happens if nobody answers?
Do I need to turn anything on?
Run tools that ask questions from any agent, even one that can't show the form
Run any MCP server in Meandr or connect the ones you already have, then point every agent at one MCP gateway. Policy, approvals and audit cover every call, with no code changes.
Try for freeQuestions? Email us