MCP Tool Not Working in an AI Coding Workflow

When a run goes wrong, don't re-prompt and hope. Narrow the problem to one symptom, 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. A run that went badly can still contain an edit or two that were right, and if you throw everything away you pay for that work again. The diff is also easier to read than the transcript when you are working out where the run went wrong.

Re-prompting harder almost never works. If the agent misread the task, adding "please be careful" to the same prompt gives it nothing new. New information changes the outcome: the file it never opened, the exact error text instead of your paraphrase, or a boundary tight enough to remove the ambiguity. If the second attempt fails the same way as the first, look at the prompt before you blame the model.

Most of these failures trace back to a gap the run never closed: the agent missed the context that mattered, the prompt set no boundary, or nothing checked the result before it landed. Name the gap before the next run, because each one needs a different fix. Adding context, for example, does nothing for a run that had no boundary.

These two failures look the same from the outside, and they need different fixes. If the server never started, no tools appear at all. Open 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), which show server initialization, tool calls and error messages. If the tools are listed and enabled and the run still ignored them, the config is not your problem, and reloading won't help.

In that second case, check how your run mode treats 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. calls.

In Allowlist mode, Cursor asks for approval before a tool that isn't on the allowlist runs, so a run you weren't watching may be waiting on a prompt. Auto-review, the default from Cursor 3An older release label for Cursor's multi-agent interface; current product guidance calls the active surface the Agents Window. Press Enter for the full definition..6, sends those calls to a classifier instead. If the classifier blocks a call, the agent can try another approach, which looks exactly like an agent ignoring your tool.

Then count your tools. With many tools enabled across servers, the agent can pick the wrong one, so turn off what this task doesn't need. To see what your servers cost, click the context ring next to the prompt input. Its breakdown shows the tokens spent on tool definitions and on the instructions and catalogs of connected 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.

You can also name the tool you want. A follow-up that names 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. server can steer a wandering agent back to it mid-run.

Because that run happens on a different machine with a different list of tools. A cloud run gets your repo, its in-repo rules and its project skills, but its 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 the ones configured for your team and the personal servers you enable for cloud agents. So first check whether the server was available to that run at all.

Sign-ins are separate too. Cloud AgentsAgents that run in a Cursor-managed virtual machine, check out the repo, do the work and open a pull request, then shut down, with no load on your laptop. Press Enter for the full definition. keep their own 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. sign-ins on Cursor's servers, and when a server's sign-in fails, the run skips that server's tools and carries on.

An automation is narrower again. 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. is one of the optional tools you choose for each automation, so check that automation's tool list before you change anything on the server.

Some failures are about the network, not the configuration. A server that only exists inside your network has to be reachable from wherever the run executes. Cursor's documented option is Self-Hosted Machines: a machine you operate runs the agent's tools and connects out to Cursor, so a command (stdio) 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. server runs on that worker and can reach private endpoints.

A remote 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. server that Cursor reaches by URL is a different case. Cursor's backend still calls that URL, even when the run uses a self-hosted worker, so the URL has to be reachable from Cursor's backend, not just from your machines. Rule both cases in or out before you spend an afternoon on mcp.json.

Don't expect 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. server to enforce anything either. Cursor's docs say MCP servers provide information rather than enforce policy, and MCP tools can still read files your .cursorignore excludes. The credential you give a server is the real boundary. Setup details for both transports are in the MCP setup guide.