MCP server

DailyBlogPost exposes a remote MCP server (Model Context Protocol) so your own AI agent — Claude, Cursor, or any MCP-compatible client — can drive your blog end to end: read your measured SEO signals, decide what to write, create topics, launch generations and publish, all through the same guardrails as the app.

Nothing to install. It is a remote server:

https://dailyblogpost.app/api/mcp

Authentication is your organization API token, sent as a Bearer token.

Quick start

1. In Settings → API, create an API token. Read-only scopes are pre-selected — tick the write scopes (topics:write, generate:write, publish:write) only if you want your agent to act, not just observe.
2. Add the server to your MCP client.

Claude Code

claude mcp add --transport http dailyblogpost \
  https://dailyblogpost.app/api/mcp \
  --header "Authorization: Bearer $DBP_API_TOKEN"

Claude (claude.ai / Desktop) — add a custom connector with the URL above and your token as the Bearer/authorization value.

Cursor — add to .cursor/mcp.json:

{
  "mcpServers": {
    "dailyblogpost": {
      "url": "https://dailyblogpost.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_DBP_API_TOKEN"
      }
    }
  }
}

3. Ask your agent something like: "Look at my DailyBlogPost SEO signals and propose the three articles with the best chance of ranking."

Scopes

Each tool requires a scope carried by your token. The organization is always derived from the token — no tool ever takes an organization parameter.

Write scopes are never pre-selected when creating a token: giving your agent the power to act is an explicit choice. You can revoke a token at any time in Settings → API — revocation is immediate.

Tools

Read — the measure half of the loop

Write — the act half of the loop

publish_article goes through the same publication checkpoint as the app. Articles that are not ready are refused outright — citations that cannot be verified against the research sources, missing SEO fields, a language that does not match the site, a quality score under your organization's threshold. These refusals are not overridable. One class of confirmable warnings exists (e.g. an article with no persisted sources): the tool returns the warning and publishes only if called again with acknowledgeWarnings: true.

What your agent can do with this

The tools are designed to close the loop, not just expose data:

• Weekly editorial review — "Check my SEO signals, summarize what moved, and propose next week's articles from the untargeted queries."
• Opportunity to article — "Take the best untargeted opportunity, create a topic for it, generate the article and tell me when it is ready for review."
• Guarded autopilot — "Every Monday, generate one article from my best opportunity and publish it if it passes all checks; otherwise leave it in my review queue and tell me why."

Your quotas (topics, monthly articles) and your subscription are enforced on every call — they are also the natural budget ceiling for an agent running in a loop.

Security model

• Organization from the token, always. No tool accepts an organization or account identifier; your token scopes everything it can see and do.
• Scopes per tool, checked on every call.
• Rate limiting per token, plus an audit journal: every tool call is logged with its tool name, outcome and duration.
• Bounded outputs, no secrets. Responses never contain integration credentials, tokens or emails. Search queries returned by the signals tools are external text typed by real users — your agent should treat them as data, never as instructions.

Troubleshooting

• 401 — missing, mistyped or revoked token. Recreate one in Settings → API.
• "This token is missing the required scope" — the tool needs a scope your token does not carry. Create a token with the right scopes (scopes are fixed at creation).
• Rate limit reached — the response includes the reset time; space out the calls.
• generate_article says the topic is no longer available — the topic was already generated (one article per topic). Check list_articles.