Ask a question
How do I…? Is this expected? Help with setup. Answered in GitHub Discussions.
Report a bug
Something is broken or behaves differently from these docs.
Suggest a feature
A change or addition you’d like. Early ideas can start in Discussions → Ideas.
Report a security problem
Privately, never in a public issue or discussion.
From inside AgentSDR
Open Help in the sidebar. It links to these docs, the bug and feature forms, Discussions and private security reporting. Report a bug fills in the version you are running, so you don’t have to look it up.
The Help menu in the sidebar.
Before you ask
1
Search the docs
Use the search at the top of this page. Setup problems are usually covered in Self-hosting troubleshooting or the Troubleshooting section of the integration’s guide.
2
Search existing issues and discussions
Look through issues (open and closed) and Discussions. If someone has already reported it, add a 👍 or your details there instead of opening a new one.
3
Remove secrets and personal data
Issues and discussions are public. Before you paste logs,
.env lines or screenshots, take out API keys, passwords, database URLs, and your leads’ names, emails and phone numbers.What makes a bug report easy to fix
The bug form asks for each of these:- Version or commit. The version from Help → Report a bug is filled in for you. On a source install,
git rev-parse --short HEADgives the commit. - How you run AgentSDR: Docker Compose, from source, or local development.
- Steps to reproduce, what you expected, and what happened instead.
- Log lines from around the failure:
docker compose logs app --tail 200on Docker Compose, or your process manager’s output on a source install. - The area, for example email, LinkedIn, WhatsApp, Tables or self-hosting.
Using AgentSDR from an AI assistant
Can I use AgentSDR from Claude?
Can I use AgentSDR from Claude?
Yes. AgentSDR is an MCP server, so Claude (web and desktop), Claude Code, Cursor and other MCP clients can connect and work the Action required queue. Setup is in AI assistants (MCP), and Settings, AI assistants shows your server URL.
Can the assistant send without asking me?
Can the assistant send without asking me?
Only if you allow it twice over. Sending needs the Send messages permission (
crm:send) when you connect, and even then the server tells the assistant to show you the exact text and wait for a yes unless you told it to send without asking. Untick Send messages if you never want it to send.How do I disconnect an assistant?
How do I disconnect an assistant?
Open Settings, AI assistants, Your connections and choose Disconnect. It stops at once, including tokens it already holds.
Claude says it couldn't determine how the server signs in
Claude says it couldn't determine how the server signs in
See Troubleshooting on the AI assistants page.
Security problems
Don’t describe a vulnerability in an issue, a discussion or a pull request. Report it privately through GitHub private vulnerability reporting. Only the maintainers can see the report. Include the affected version, the component, and the steps to reproduce. The full policy is in SECURITY.md. If you notice a credential or personal data posted in public by mistake, report it the same way so a maintainer can remove it.What happens after you report
- A maintainer reads every new issue. It starts with the
needs-triagelabel, which comes off once someone has looked at it. - If something is missing, they’ll ask for it in the issue. Reports that stay incomplete may be closed.
- A confirmed bug gets the
buglabel. Small, well-defined ones may be labelledgood first issuefor new contributors. - When a fix lands, the issue closes automatically and the fix is listed in the changelog of the release that includes it. Update your install to get it (Upgrading).
Contribute a fix
If you’d like to fix it yourself, start with Development for local setup and conventions, then read CONTRIBUTING.md. Issues labelledgood first issue are a good place to start.