Model Context Protocol
Connect Claude to your product feedback
Reading four hundred posts is a job for a machine. SignlOS speaks MCP — the open standard for connecting AI assistants to real systems — so you can ask what your customers want and get an answer grounded in your actual boards and votes.
Endpoint
POST https://api.signlos.com/mcp
Authorization: Bearer sk_live_…In practice
Questions a filter cannot answer
A feedback board gives you sorting and tags. What you usually want is a judgement — which of these requests are the same thing, what are enterprise accounts blocked on, what did we ship without telling anyone. Those are questions, not filters.
The assistant reads the real data through the same permissions your team has, so the answer cites specific posts and vote counts rather than paraphrasing a summary somebody wrote in March.
- What are the top ten requests we haven't planned yet?
- Summarise everything customers said about billing in the last 90 days.
- Which roadmap items shipped but have no release note?
- Are there duplicates of post #412 I should merge?
- Draft a release note for everything we moved to Shipped this week.
- Which enterprise accounts are waiting on something we haven't started?
Reference
9 tools
Each tool checks the same scopes as the equivalent REST endpoint. A key without a scope does not see the tools that need it.
| Tool | What it does | Scope | Writes |
|---|---|---|---|
| list_boards | Every board in the workspace, with post counts. | boards:read | No |
| search_posts | Search and filter posts by text, board, status, tag, or vote count. | posts:read | No |
| get_post | One post in full — description, votes, requesters, comments. | posts:read | No |
| feedback_summary | What the board says right now: top requests, recent movement, recurring themes. | posts:read | No |
| get_roadmap | Roadmap items by column, with the feedback posts feeding each. | roadmap:read | No |
| list_changelog | Published release notes, newest first. | changelog:read | No |
| get_status | Component health and any open incidents. | status:read | No |
| create_post | File a new post on a board. | posts:write | Yes |
| update_post_status | Move a post — which notifies everyone who voted for it. | posts:write | Yes |
Security
The key is the boundary
There is no separate MCP credential to issue, audit, or forget about.
One credential
The assistant uses your existing scoped API key. Revoking the key revokes MCP access in the same moment — there is no second place to check.
Scopes are enforced server-side
A read-only key cannot write, no matter what the assistant is asked to do. The check happens in SignlOS, not in the prompt.
One workspace, always
A key resolves to exactly one workspace and every query is scoped by its tenant id. It cannot address another workspace, whatever it asks for.
Writes are visible to customers
update_post_status emails everyone who voted on the post. It is not a reversible sandbox action. Start with a read-only key and add write scopes only once you want the assistant acting on your behalf in public.
Setup
Three minutes, two config lines
Claude Code
claude mcp add --transport http signlos https://api.signlos.com/mcp \
--header "Authorization: Bearer sk_live_…"Claude Desktop — claude_desktop_config.json
{
"mcpServers": {
"signlos": {
"type": "http",
"url": "https://api.signlos.com/mcp",
"headers": {
"Authorization": "Bearer sk_live_…"
}
}
}
}Full walkthrough, including which scopes to grant: Connect Claude to SignlOS.
Questions
Frequently asked
Your customers already told you what to build
It is in a support inbox, three Slack threads, and a spreadsheet nobody has opened since March. Put it somewhere it can be counted.
No credit card required to start. $0 forever.