Niche for developers

Build agents on the
editorial intelligence MCP.

20 workflow-grained MCP tools — signal → angle → draft → render → publish — with verifier-checked claims and source provenance on every output. Install on Claude Desktop, Claude Code, Cursor, or any MCP-compatible client in 5 minutes.

Five-minute install

1. Generate a PAT at /settings/api-keys. Tokens prefix with niche_sk_.

2. Add the server to your agent client. Claude Desktop (~/.config/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "niche": {
      "transport": "streamable-http",
      "url": "https://api.nicheangle.com/mcp",
      "headers": {
        "Authorization": "Bearer niche_sk_..."
      }
    }
  }
}

Claude Code (CLI):

claude mcp add niche \
  --transport streamable-http \
  --url https://api.nicheangle.com/mcp \
  --header "Authorization: Bearer niche_sk_..."

Cursor (Settings → MCP → Add server): same shape as Claude Desktop's JSON.

3. Restart the client. Your agent now sees 20 Niche tools via tools/list.

The 20-tool surface

Tools group into 6 functional bands. Full schemas live at /.well-known/agent-skills/index.json (auto-generated from the live registry — never drifts from the server's tools/list response).

Discovery + state

5 TOOLS

Start an editorial run for a niche, poll/long-poll session state with `since_status` (saves agent tokens), revert to any prior checkpoint, cancel mid-flight.

niche_signal_scanniche_session_stateniche_list_sessionsniche_session_revertniche_session_cancel

Editorial

2 TOOLS

Propose 3 angles per picked story with frame / hook / tension / CTA + verifier audit. Cross-angle exploration scores OTHER ranked stories against the chosen frame — primitive nobody else has.

niche_angle_proposeniche_angle_explore

Drafting

3 TOOLS

Generate platform-native drafts with trust fields (faithfulness score, ungrounded claims, source diversity, recency). Sub-field revision with diff[] return. Post-CP3 add-on of new output cells.

niche_draft_createniche_draft_reviseniche_add_output

Rendering

2 TOOLS

On-demand image card + 9:16 reel with cell-aware dimensions (1200×627 for linkedin:image_post, 1080×1350 for instagram:image_post, etc.). Idempotent replace.

niche_render_image_cardniche_render_reel

Brand intelligence

7 TOOLS

Persisted BrandProfile JSON with schema introspection via `_template`. Ingest brand from URL / files / freeform text. Voice extraction from past posts. Guided-setup question chain for first-touch onboarding.

niche_brand_profile_getniche_brand_profile_setniche_brand_profile_templateniche_brand_kit_ingestniche_brand_kit_updateniche_brand_kit_guided_setupniche_voice_profile_ingest

Publishing

1 TOOL

Dry-run by default. Idempotency-key required for commit. Verifier-blocked outputs refuse to publish even at `dry_run=false`.

niche_draft_publish

Trust fields on every draft

Every output row carries the trust block — agents can gate publish decisions on these without re-deriving them:

verifier_blocked_reason + verifier_blocked_claim — present when a load-bearing fabrication couldn't be grounded. Publish refuses even at dry_run=false.
source_faithfulness_score ∈ [0.0, 1.0] — fraction of concrete claims in the draft that appear verbatim or normalized in the source material.
source_ungrounded_claims — list of specific tokens (numbers, dates, proper nouns) the verifier couldn’t ground. Agents typically gate at >0 entries.
source_diversity_passed + source_recency_passed — boolean checks on the contributing signal pool.

Endpoints

MCPapi.nicheangle.com/mcp · Streamable HTTP · Bearer PATOpenAPIapi.nicheangle.com/openapi.jsonAgent skills/.well-known/agent-skills/index.jsonAPI catalog/.well-known/api-catalog (RFC 9727)

Going deeper

The full walkthrough — brand-kit setup, voice profile, sample end-to-end agent run, troubleshooting — lives at /integrations. That page covers both the human web app and the agent surface; this page is the developer-shaped subset.