Tool-call limit in MCP causing unwanted pauses

My second post on this subject in the last couple of months, See here:

https://feedback.boltai.com/p/constant-halts-using-mcp-filedirectory-readswritessearches-per-this

Problem:
I give the AI a task involving let's say refactoring the MCP server structure or some task involving copying or searching requiring a collection of MCP tool calls. It executes at most four or five tool calls and then pauses with no user feedback. I have to issue a character or a question then hit command return to get it to continue. Then another pause after four tool calls. This behaviour makes it quite tedious to use BoltAI compared to an online experience with Claude or Gemini. See the AI’s interpretation of the issue in the form of a recent chat:

Me:

Why do you pause with no feedback after every four or so MCP tool calls? This keeps occurring, there must be a reason. I have called out this behaviour many times but no fix seems to work despite AI's promises.

AI:

The host environment enforces a pause after a set number of consecutive tool calls. This mechanism interrupts autonomous execution to prevent infinite loops and runaway token consumption. Git version control allows for code regression; however, the tool-call limit remains the primary method to restrict unbounded token usage.

ME:

Can we release the host environment from enforcing a pause after a set number of consecutive tool calls?

AI:

The AI cannot modify or release the host environment's tool-call limit. The client application enforces this pause as a hardcoded safety mechanism specifically to prevent the runaway token consumption noted previously.

Me:

Is that limit part of bolt AI, or is it part of my preferences setup for the AI? explain in detail

AI:

The tool-call limit is enforced by the client application, not the Gemini AI preferences.

𝐂𝐥𝐢𝐞𝐧𝐭 𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐌𝐞𝐜𝐡𝐚𝐧𝐢𝐬𝐦
• The client application orchestrates the execution loop. It receives tool call requests from the model, executes them via the MCP server, and returns the results to the model.
• To prevent infinite execution loops and runaway token consumption, the client application hardcodes a maximum number of consecutive autonomous iterations.
• When this threshold is reached, the client application suspends the execution loop and requires explicit user approval to resume. This is a safety boundary built into the host environment.

𝐌𝐨𝐝𝐞𝐥 𝐏𝐫𝐨𝐯𝐢𝐝𝐞𝐫 𝐌𝐞𝐜𝐡𝐚𝐧𝐢𝐬𝐦
• The Gemini AI model processes individual prompts and generates responses. It does not manage the overarching execution loop.
• The model provider enforces limits on context window size and maximum output tokens per request. It does not track or restrict the number of consecutive tool calls executed by the client application.
• Adjusting Gemini AI preferences will not override the client application's hardcoded execution limits.