All notable changes to the CitationBench REST API and MCP server. New endpoints, breaking changes with 12-month deprecation windows, bug fixes, and model upgrades.

All notable changes to the CitationBench API and MCP server. New endpoints, breaking changes (with deprecation windows), bug fixes worth noting, and model upgrades.

The changelog is also available as RSS at https://citationbench.com/changelog.rss.

Stability policy

/v1/ is stable. Breaking changes follow a 12-month deprecation window with Sunset headers and changelog entries.
Beta endpoints live under /v1/beta/* and may change without notice. They graduate to /v1/* when stable.
The MCP tool surface tracks REST stability. Tool renames respect the same deprecation window.

2026-05-24 — `v1.0.0` (initial)

Public launch of the CitationBench /v1/ API + MCP server.

Pillars

Agent — one generic agent + skills registry. agent.invoke, agent.files, agent.approval. ~15 endpoints.
Research — keyword, serp, serp gap analysis, competitor, icp, discuss, ai citation, content gap. ~30 endpoints.
Production — blog post, landing page, refine, evaluate (new), image, publish. ~36 endpoints.
Indexing — gsc index, indexnow. ~13 endpoints.
Link Building — serp outreach, competitor link outreach, inbound (with eval gates), campaign management, CRM. ~28 endpoints.

New endpoints not seen elsewhere

research.serp_gap — detect SERP cliffs to find winnable keywords
produce.evaluate — rubric-based content scoring
link_building.inbound — autonomous negotiator for inbound link-building requests, with eval gates

Authentication

API keys: sk_live_*** (prod), sk_test_*** (sandbox)
Demo mode — every endpoint responds without auth, returning shape-complete synthetic data. Lets AI agents integrate before the user creates an account.

Skills (built-in)

bootstrap_brand — 20-minute new-brand setup
rank_monitor — scheduled rank checks with auto-remediation hooks
link_hunter — end-to-end SERP + outreach
citation_hunter — daily AI search visibility tracking + reclamation
content_factory — keyword → research → draft → refine → publish
refresh_stale — rank/citation drop → refresh draft
keyword_manager — conversational keyword DB ops
link_swap_evaluator — score partner link-swap offers
inbound_negotiator — autonomous inbound response

MCP server

Hosted at mcp.citationbench.com/mcp
Supported clients: Claude Code, Claude Desktop, Cursor, Windsurf, ChatGPT Apps, custom (HTTP transport)
~131 MCP tools mirroring the REST surface 1:1

Concepts

One generic agent + skills (not many named agents)
2D keyword labeling (intent × relevance)
Content refiners (SIMPLE, IMAGE, MULTI_IMAGE)
Workspaces (per-brand isolation, agency-wide bulk ops)
Approval workflows (workspace + skill + per-invocation policy layers)
Eval gates (conditional escalation rules)

Webhooks

30+ event types across agent, research, production, indexing, link-building
HMAC-SHA256 signed
4-retry policy + dead-letter queue

Roadmap (subject to change)

Things planned for the next 6–12 months. Tracked publicly to set expectations; dates are aspirational.

Q3 2026

/v1/beta/agent.studio — workspace-scoped custom skill authoring API
Webflow direct publish (currently REST-hook only)
White-label dashboards (agency tier)

Q4 2026

Reseller pricing for agencies (sub-account billing roll-ups)
Compliance hardening: SOC2 Type 1, GDPR data residency in EU
/v1/beta/agent.simulator — replay an invocation against demo data to verify changes

2027 (early)

/v1/beta/produce.video — short-form video generation from blog posts
/v1/beta/research.serp.history — full historical SERP archive (currently 90 days)
Multi-agent runtime — opt-in for users who want multiple distinct agents per workspace

RSS: https://citationbench.com/changelog.rss
Email digest: opt-in at Settings → Notifications
Webhook: subscribe to changelog.entry.created for programmatic updates

Changelog — CitationBench API & MCP server release notes