The Model Context Protocol (MCP) gives an AI client a standard way to discover tools and call them through a server. Signalstead’s remote MCP endpoint exposes read-only tools for one authorized workspace, including saved audit findings, Search Console query summaries and keyword history with limited query-to-page examples, approved facts, draft and release status, AI visibility observations, and AI answer accuracy totals.
Create a project-specific connection
- Sign in to Signalstead and select the workspace you want the assistant to access.
- Open Workspace settings → AI assistant connection (MCP).
- Name the connection and choose Create connection.
- Copy the configuration while the token is shown. Signalstead stores a hash and will not display the token again.
- Paste the configuration into an MCP client that accepts a remote Streamable HTTP URL and custom authorization headers.
Each credential belongs to the creating user and selected workspace. It expires after 90 days, can be revoked from settings, and is limited to 30 requests per minute and 200 requests per day. Create a separate credential after switching to another workspace.
What the assistant can read
The available tools can inspect a workspace overview, saved audit findings, Search Console query performance and saved keyword snapshots with limited query-to-page examples, approved business facts, saved draft status, recorded release checks, AI answer accuracy totals, stored AI visibility observations, recommendations, and supporting evidence. Keyword snapshots include a period comparison when an earlier saved window is available; the windows are rolling 90-day totals and may omit rows. They can also prepare evidence-backed content outlines in the client response. MCP cannot create or change drafts, approve facts, publish website changes, or trigger a Signalstead AI-provider run. Pending facts and saved AI answer text are not returned by the business-facts and accuracy-summary tools.
An on-demand competitor review crawls at most five public pages, respects robots.txt, and compares their visible page coverage with your latest saved site crawl and your own Search Console query data. It is limited to three crawls per workspace user each hour and is also available in the app under Competitors → Insights from AI answers. Competitor Search Console and analytics figures are private, so Signalstead cannot retrieve them. Page gaps are review hypotheses, not proof of better performance.
Reads and competitor page reviews use no Signalstead AI credits or model calls. The bounded public crawl does use server requests; the connected client may charge for its own model interpretation. Workspace settings shows recent per-connection request use and the shared crawl limit.
Connect with OAuth
OAuth-enabled remote clients can discover Signalstead’s authorization server, register as a public client, and ask you to sign in and approve read-only access. The authorization is scoped to the workspace selected in your Signalstead session. For a successful connection, the client should discover the server, complete authorization, initialize the MCP session, and list the available tools. Anthropic documents remote MCP OAuth and Dynamic Client Registration in its connector guide.
View or revoke OAuth clients and bearer-token connections in Workspace settings → AI assistant connection (MCP). OAuth clients receive short-lived access tokens with rotating refresh tokens; the connection stays listed while it is active.
Keep the token safe
Treat the credential like a password. Do not paste it into a prompt, public issue, screenshot, or shared document. If it is exposed, revoke it in Workspace settings and create a replacement. MCP access remains limited to the workspace attached to that token; never put a customer token into a public example.
See the Signalstead MCP setup walkthrough for the current connection instructions.
Primary sources: Anthropic: Building custom connectors via remote MCP servers · Model Context Protocol: Streamable HTTP transport.