Fix MCP, cloud and Jira integrations
Find the cause and the fix when a registered MCP server is not available, an AWS or Google Cloud command stops, jira ends with an error, or an app connected from outside shows no tools.
An MCP server you registered is not available in sessions or assistants
Section titled “An MCP server you registered is not available in sessions or assistants”Work through ⚙ Settings → MCP servers (used by sessions) from the top.
- Is the server enabled?
- Do the targets under Advanced include sessions or assistants, and are the target agents not narrowed too far?
- Does it say “These personal registrations are unused because a tenant-distributed server has the same name: …”? The tenant’s server of the same name wins
- Press “Test connection”: do a server name and a tool count come back?
A change applies to the sessions you start next. A running session does not change until it restarts. Missing values, and asking for outbound access, are in Add an MCP server to your sessions.
The connection test passes, but the session says Session not found
Section titled “The connection test passes, but the session says Session not found”This happens when a server that keeps MCP sessions in memory (a stateful Streamable HTTP server) runs as several instances without session affinity. One instance issues the session ID, and the next request lands on another that has never heard of it. When the connection test finds this, it says “The server rejected the session ID it just issued. …”.
Agent Fleet cannot fix it on its side. Change the server to one of these:
- Make it stateless (a stateless Streamable HTTP mode, for example)
- Run a single instance
- Route every request with the same
Mcp-Session-Idto the same instance
Load-balancer cookie stickiness does not help, because Claude Code does not send cookies back on MCP requests. Once the server is fixed, restart the session. The connection test is described in Add an MCP server to your sessions.
An MCP server’s row says “Sign in again” or “More permission needed”
Section titled “An MCP server’s row says “Sign in again” or “More permission needed””These are the states of an OAuth server’s row. An OAuth server is handed to sessions and assistants only while you are signed in.
| The row says | What it means | What to do |
|---|---|---|
| “Sign in again” | The saved sign-in can no longer be renewed | Press “Sign in again”. Until you do, sessions you start are not given the server |
| “More permission needed” | You have not yet granted a permission a tool needs | Press “Approve the extra permission”. It signs you in again, adding that permission to what you granted |
| A note that the sign-in has not finished | A sign-in was started and never came back | Retry as described in Add an MCP server to your sessions |
An app connected from outside shows no tools
Section titled “An app connected from outside shows no tools”In ⚙ Settings → MCP tokens & apps, open “Recent calls” for that app under “Connected apps”.
Check the HTTP status of its requests and how many tools tools/list returned.
- If the list looks empty, refresh it
- If needed, connect again from the app
The details are in Connect a remote MCP client by signing in.
Jira says boards, sprints and epics are “not available”
Section titled “Jira says boards, sprints and epics are “not available””The OAuth connection lacks the Jira Software scopes. The Jira card in ⚙ Settings → Issue tracker says “Boards, sprints and epics: not available. …”, and jira init and the board commands fail with an authorization error.
- Ask your tenant administrator to add the scopes to the Atlassian app and tick the Jira Software option on the tenant’s Jira card
- Then reconnect Jira once
Retrying jira init before that changes nothing. Viewing, listing, creating and commenting on issues do not need these scopes (Use Jira).
jira issue create or comment add exited with status 124
Section titled “jira issue create or comment add exited with status 124”The wrapper cut the command off at its time limit. The issue or comment may already exist, so do not run it again straight away. Check in Jira first, and create it again only if it is not there. How to check is in Use Jira.
jira exited with status 3
Section titled “jira exited with status 3”jira issue create or jira issue edit -b could not set the description, and nothing else failed.
After a create the issue exists, so do not create it again. Set the description with jira issue edit (Use Jira).
An AWS command says “Unable to locate credentials”
Section titled “An AWS command says “Unable to locate credentials””The command found no credentials of yours. The workspace’s own permissions are not handed to your shell.
Do not configure credentials: use a profile from ⚙ Settings and run the command through af-aws-exec.
af-aws-exec --profile <name> --account <id> -- <command>The fix is in Use AWS SSO, and every option is in af-aws-exec.
af-gcloud-exec exited with status 3
Section titled “af-gcloud-exec exited with status 3”Status 3 means a login only when the wrapper says so (“the login was requested in the Agent Fleet Console …”, “Google Cloud login required …”). Log in from the toast or ⚙ Settings, or run the terminal command it printed, then run the command again. Otherwise the status is the command’s own: read that command’s error (Use Google Cloud).