MCP limits and troubleshooting
Limits
Section titled “Limits”| Limit | |
|---|---|
| Requests | 240 a minute for one person, 600 a minute for one workspace (RATE_LIMITED above) |
context_upload |
600 files and 4 MB per call; text files only, files over 1 MB left out |
context_read |
200 lines per call |
decide |
Waits up to a minute; then read the outcome with decision_get |
| A task claim | 10 minutes from the last task_heartbeat or task_report |
| A sign-in | 30 days without use, 90 days at most |
| Coding tools that can register | 1,000 for the whole service |
Error codes
Section titled “Error codes”Every refused call returns a message that starts with a code and ends with the request id: CODE: what happened (request req_…). Quote the request id when you write to info@weharness.io.
| Code | What it means | What to do |
|---|---|---|
AUTH_REQUIRED |
The tool is not signed in, its sign-in expired or was disconnected | Sign in again in the tool (for example /mcp in Claude Code, codex mcp login weharness) |
NOT_FOUND |
The workspace, project or item does not exist, or you are not a member | Check the ids with project_inspect; ask the workspace’s owner for an invitation |
FORBIDDEN |
Your role does not allow this | Ask an owner or admin, or have them do it |
INVALID_ARGUMENT, VALIDATION |
A parameter is missing or has the wrong shape, for example an id that is not a UUID | See the tool’s parameters in the reference |
RATE_LIMITED |
Too many requests in a minute | Wait a minute |
OUT_OF_CREDITS |
The workspace has no credits for new billable work | Running work finishes. Ask WeHarness to add credits. See Credits and usage |
MEMBER_BUDGET_REACHED, RUN_BUDGET |
Your monthly budget is used, or the work would reserve more than your per-run budget | Ask an owner or admin to change your budget in Usage & billing |
WORKSPACE_BUSY, AT_CAPACITY, PROVIDER_BUSY |
Too much work is running in the workspace or in WeHarness | Try again in a minute; the answer says when |
WORK_PAUSED |
New work is paused by an automatic safety measure, with a reason code | Work already running finishes. If it looks wrong, write to info@weharness.io with the reason code |
APPROVAL_REQUIRES_SESSION, WEB_SESSION_REQUIRED |
The action is an approval, which only a person in the web app can make | Open the approve_in or decide_in link in your browser |
Sign-in problems
Section titled “Sign-in problems”The browser does not open, or sign-in never finishes. Run the tool’s sign-in command again (see your tool’s page under Connect your coding tool). Your tool must be able to receive the redirect on localhost or 127.0.0.1: a firewall or a remote development container can block it.
“Invalid redirect” or registration refused. Registration accepts redirects only to your own machine (localhost or 127.0.0.1, any port) and the known callbacks of a few web tools. A tool that runs in someone else’s cloud, such as Claude on the web or ChatGPT, cannot register yet. See Claude web and ChatGPT.
The token is refused right after sign-in. Your email address must be verified. Sign in to app.weharness.io once and follow the verification link if WeHarness sign-in asks for it.
It worked yesterday and now asks me to sign in. Its sign-in expired (30 days without use, 90 days at most), someone used Disconnect or Sign out everywhere, or WeHarness staff ended MCP access for your account. Sign in again.
Connection problems
Section titled “Connection problems”The tool does not appear in Connect MCP. It appears when it first connects after sign-in. Start the tool, or ask it: “List my WeHarness workspaces.” Connect MCP refreshes every few seconds while it is open.
A tool is missing from the list my coding tool shows. Your tool may cache the tool list. Restart it, or reconnect the server. The current list is in the reference and the Server Card.
approve_in opens a page that says not found. Sign in to the web app with the same account your tool uses, and choose the right workspace.
My tool says its checks passed, but the report failed. Host-reported checks and independent cloud reproduction are different evidence. Read task_report_get for the latest candidate and task_next for missing checks or the next step. Local reports do not start a sandbox automatically. See Working on an approved plan.
Tool evidence export (next deployment)
Section titled “Tool evidence export (next deployment)”The release with migration 0103 records personal MCP tool receipts for fourteen days.
After that release is deployed, sign in to the app and open /api/v1/me/mcp/evidence.
Follow both fields in the next cursor to retrieve subsequent pages.
Receipts include timing, returned/error state, structured output facts, artifact IDs and the request id a tool error names; they do not include raw prompts, code, credentials, exception text or your coding tool’s chat. The export holds only your own receipts. A returned call does not prove the advice was useful or that the host applied it. Open referenced plans and reports with your normal project permissions, and compare against the same task without WeHarness.
In the web app, Build → WeHarness work shows an application’s calls next to its plans, tasks, evaluation runs and other WeHarness runs, newest first, apart from the application’s traces. You always see your own coding tools’ calls there; the workspace’s owners and admins also see the calls of its other members on that application. Members and viewers never see another person’s calls. To find a failed call, look for the request id from the error.
Still stuck
Section titled “Still stuck”Write to info@weharness.io with your tool’s name and version, what you asked, and the request id from the error.