MCP Tool Not Working in an AI Coding Workflow

When a run goes wrong, resist the urge to re-prompt and hope. Narrow it to one symptom first, then make the smallest change that tells you whether you guessed right.

1. 1Stop the run and save the current diff.
2. 2Check whether the task was too broad or missing files.
3. 3Read the failing command, log or review comment out loud.
4. 4Start a smaller run with one clear done state.
5. 5Run the narrowest check before asking for more changes.

Save the diff before you do anything else. Even a run that went badly usually has an edit or two in it that were right, and if you throw the whole thing away you'll just pay for that work again. It's also far easier to read than the transcript when you're trying to work out where things went sideways.

The instinct is to re-prompt, harder. Look, it almost never works. If the agent has already misread the task, adding "please be careful" to the same prompt gives it precisely nothing it didn't have the first time. What actually changes the outcome is new information: the file it never opened, the error text you paraphrased instead of pasting, a boundary tight enough that the ambiguity disappears. When the second attempt fails the same way as the first, that's the prompt telling you something, not the model.

Most of these failures trace back to a gap the run never closed. Usually the agent missed the context that mattered, the prompt never set a boundary, or nothing checked the result before it landed.

Two different failures look identical from the outside, and they take opposite fixes. When the server never started, no tools appear at all, and the MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. Logs in the Output panel (⌘+⇧+U) usually name the reason. When the tools are listed and enabled and the run ignored them anyway, the config is not your problem and reloading will not help.

In that second case, check approval first. Cursor asks before it calls an MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. tool by default, so a run you were not watching can be sitting on a prompt rather than failing.

Then count your tools. My guess is this unglamorous one explains more "the agent won't use my MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition." reports than any config error does. With too many tools enabled across servers, the agent picks badly, so trim to the ones this task actually needs. Cursor's context usage report breaks down where tokens go across the system prompt, tools, rules and skills, which is often the argument that settles how many servers stay on.

You can also just name the server you want. Telling a wandering agent to use a specific MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. mid-run pulls it back onto the tool it should have reached for.

Because that run is a different machine with a different tool list. A cloud run mirrors your repo, your in-repo rules and your skills, and MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. servers are linked in the launch flow, so the first thing to check is whether the server was linked to that run at all.

An automation is stricter again by design. Its tools are the actions it is allowed to take, scoped to least privilege, so the fix is usually attaching the tool to that automation rather than anything to do with the server.

Then there is the case that is not a configuration problem at all. If the server only exists inside your network, a run in Cursor's cloud cannot reach it, and the documented path is a self-hosted worker: the VM runs in your network so it can reach on-prem source control and internal MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. servers. Worth ruling that in or out before you spend an afternoon on mcp.json.

And do not expect the server to enforce anything. MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. tools inform rather than block, and they cannot apply your .cursorignore, so the credential you hand one is the real boundary. Setup detail for both transports lives in the MCP setup guide.