Connect over MCP

Point Claude or any MCP client at Inklate's Streamable HTTP server, authenticate with an API key or OAuth, and call the full tool catalog.

Inklate exposes a Model Context Protocol server so any AI agent — Claude Code, Claude Desktop, Cursor, and other MCP clients — can discover channels, author posts, preflight, schedule, publish, and measure through natural language. The agent drives the same procedures the dashboard uses, so an agent and a human can never diverge.

Endpoint and transport

Transport: Streamable HTTP.
Production URL: https://api.inklate.com/mcp
Local dev: http://localhost:3000/mcp (the server runs on :3000, the web dashboard on :3001).
Auth: a bearer token in the Authorization header — either an Organization API key (prefix inklate_) or an OAuth 2.1 access token.

Scheduling and publishing are durable, background operations. create_post and preflight_post work on their own, but a schedule_post or publish_post call fails loudly rather than silently if the durable-execution layer isn’t running — a scheduled ship is never dropped.

Authenticate

Option A — Organization API key

Create a key in the dashboard under Settings → API (see API keys), then pass it as a bearer token. A key is bound to exactly one organization, so tools act on that organization with no further selection.

Option B — OAuth 2.1

Instead of a pasted key, a client can sign the user in through the browser — Inklate is a full OAuth 2.1 provider with dynamic client registration, PKCE, and a consent screen. Connect the server with no auth header; the first unauthenticated call returns 401 with a WWW-Authenticate header that kicks off discovery and the browser handshake.

An OAuth connection is not pinned to one organization. Tools act on your active organization by default; pass an organization slug argument (from list_organizations) to target a specific one. Every tool accepts this optional organization argument — it is needed only on the OAuth door.

claude.ai and Claude Desktop custom connectors require a public HTTPS URL. Local CLI clients (Claude Code, MCP Inspector) can use localhost because their OAuth callback is a local port.

Client configuration

Claude Code (CLI), API key:

claude mcp add inklate https://api.inklate.com/mcp -t http \
  -H "Authorization: Bearer inklate_XXXXXXXX"
claude mcp list          # inklate should show connected

Claude Code (CLI), OAuth — add with no header; the 401 starts the flow:

claude mcp add inklate https://api.inklate.com/mcp -t http
# then in a session:  /mcp  → Authenticate → completes in the browser

Cursor / Claude Desktop (mcp.json / MCP config):

{
  "mcpServers": {
    "inklate": {
      "url": "https://api.inklate.com/mcp",
      "headers": { "Authorization": "Bearer inklate_XXXXXXXX" }
    }
  }
}

Tool catalog

The tools are grouped by what they are for. Every tool takes an optional organization slug (needed only on the OAuth door).

Discover

Tool	What it does
`list_organizations`	Organizations (brands/clients) you can act in, with slug and your role. Pass a slug as `organization` to other tools.
`list_channels`	Publishable destinations connected to the organization plus their connection health. Posts target channels by id.
`get_capabilities`	What each channel can do right now (a publish verdict, with the reason when blocked — missing scope, disconnected, unsupported) plus every provider’s manifest (text limits, metrics, `platform_options` schema).

Author

One Story fans out into per-channel Renditions. Write the Story once; each placement’s rendition is adapted from it unless you supply an explicit variants entry.

Tool	What it does
`create_post`	Create a draft. Supply `content` (a Story doc or markdown) for cross-platform authoring, or `body` for plain text; target via `placements` (channel + format) or `channel_ids`. Also takes `variants`, `idempotency_key`, per-channel `platform_options`, and ordered `media_handles`.
`update_post`	Edit a draft / scheduled / failed post’s `content`/`body`, `placements`/`channel_ids`, or authored `variants`. Changing targets re-materializes renditions; a content change after approval re-enters the approval pipeline. Published posts are immutable.
`preflight_post`	Fit-check a draft with no persistence: per-placement publishability (reachability → scopes → content) plus adapt warnings and lengths. Run before schedule/publish.
`apply_fix`	Apply a machine-readable repair from `preflight_post`’s `refusal.fixes` to one placement’s rendition: `trim_to_fold` (over-length text), `rebalance_thread` (too many/long items), or `crop_to` (image aspect). Returns the refreshed verdict.
`sync_variants`	Copy from the Story (default) or a Rendition into target Renditions, scoped `copy` / `assets` / both. Overwrites — check `wasEdited` on the response before re-syncing edited targets.

Media

Tool	What it does
`upload_media`	Upload an image by public https `url` (preferred) or inline `base64`; returns a `med_…` handle in the `uploaded` state. Images only for now.
`get_media`	Poll an asset’s status (`uploaded → processing → ready \| rejected \| failed`), dimensions, scan status, and variants — before attaching or publishing.

Schedule and publish

Tool	What it does
`schedule_post`	Schedule (or reschedule) the whole post for a future instant; the pipeline publishes it automatically.
`schedule_placements`	Per-placement scheduling: a default fire time all placements fall back to, plus per-placement overrides, plus an optional cascade that staggers them N minutes apart.
`publish_post`	Publish immediately (also the failed-post retry path); `idempotency_key` makes retries replay. Returns a per-platform result envelope, one entry per target channel.
`unschedule_post`	Cancel a pending scheduled publish and return the post to draft.
`get_job_status`	A post’s async journey: overall status plus per-target status / external id / URL / last error — poll after publish or schedule.

Find your content

Tool	What it does
`list_posts`	Posts plus per-channel targets; filter by status (draft/scheduled/publishing/published/failed) and/or a scheduled-time range.
`search_content`	Full-text search over your own stored posts (ranked, highlighted snippets, keyset pagination, websearch syntax). Not on-platform search.
`list_interactions`	Ingested comments and mentions on your content, filtered by kind/channel/since.

Skills — durable multi-step workflows

Tool	What it does
`list_skills`	Declarative workflows you can run (e.g. `cross_platform_publish`), each with its JSON-Schema input — read it before `run_skill`.
`run_skill`	Start a skill run (async, durable) by name and input; returns a run id.
`get_skill_run`	Poll a skill run: status (running/succeeded/failed/compensated), per-step results, error.

Metrics

Tool	What it does
`get_post_metrics`	Per-channel totals (impressions, reactions, comments, shares, clicks) plus the full snapshot time series for one post.
`get_channel_metrics`	Aggregated totals plus per-day engagement across one channel’s published posts, optionally restricted to a time range.

Timestamps must be ISO-8601 with an offset — for example 2026-07-14T09:00:00-07:00 or ...Z. Naive local times are rejected with a corrective error; the agent’s client resolves the user’s timezone and sends the offset explicitly.

API keys

Create and scope an Organization API key in the dashboard.

Agent recipes

End-to-end tool sequences: draft → preflight → fix → schedule, and pull analytics.