co.pixelcove/topic-scorer

Measured conclusion: Alive, not deep-tested. L2 outcome: not_attempted.

Score pages against 11,485 public taxonomy categories by embedding similarity. $0.05/call via MCP.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "topic-scorer": {
      "url": "https://topic-scorer.pixelcove.co/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (5)

• topic_scorer_taxonomies

  Lists the six taxonomies with their label and category count. Free. Call it first to pick the taxonomy for topic_scorer_find.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• topic_scorer_find

  Finds categories whose path contains the query (case-insensitive substring), optionally within one taxonomy. Free. Returns up to `limit` hits with id and path; pass the id as `category` to the paid tools.

  Parameters:

  • query (required) — string — Words to look for in category paths.
  • limit — integer
  • taxonomy — string — Taxonomy of the target category.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• topic_scorer_score

  Scores text against every category in all six taxonomies (one embedding, cosine similarity) and returns the top `per` rows per taxonomy. With `taxonomy` and `category` it also returns the target's rank, score, gap to #1, and a phrase bank split into phrases the text already uses and phrases it does not. Read the rank and the gap, not the raw percent (scores compress into a 70 to 90 band). Paid, $0.05/call; read-only and safe to retry.

  Parameters:

  • text (required) — string — The page or draft text to score, as plain text or HTML (tags are stripped). 1 to 32768 characters. URLs are not fetched; send the content.
  • per — integer — Rows per taxonomy.
  • full — boolean — Score the whole text as the mean of 8,192-char chunks instead of only the first 8,192 chars.
  • phrases — integer — Phrase bank size for the target category (0 skips it).
  • category — string — Target category id, full path, or a unique path substring within the taxonomy (use topic_scorer_find to look one up).
  • taxonomy — string — Taxonomy of the target category.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• topic_scorer_gate

  Pass/fail check: does the text rank within the top `min_rank` categories of its target taxonomy? Returns `pass`, the target's rank and gap to #1, the taxonomy's top five, and the unused phrases to work in. Use it as a content-pipeline gate. Paid, $0.05/call; read-only and safe to retry.

  Parameters:

  • text (required) — string — The page or draft text to score, as plain text or HTML (tags are stripped). 1 to 32768 characters. URLs are not fetched; send the content.
  • taxonomy (required) — string — Taxonomy of the target category.
  • category (required) — string — Target category id, full path, or a unique path substring within the taxonomy (use topic_scorer_find to look one up).
  • full — boolean — Score the whole text as the mean of 8,192-char chunks instead of only the first 8,192 chars.
  • phrases — integer — Phrase bank size for the target category (0 skips it).
  • min_rank — integer — Highest passing rank.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• topic_scorer_phrases

  Returns the ranked phrase bank for one category: 1 to 4 word phrases a page squarely about that category would contain, ordered by similarity to the category. Use it to seed a brief. Paid, $0.05/call; read-only and safe to retry.

  Parameters:

  • taxonomy (required) — string — Taxonomy of the target category.
  • category (required) — string — Target category id, full path, or a unique path substring within the taxonomy (use topic_scorer_find to look one up).
  • n — integer — Bank size.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

Publisher website Repository

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.