# Connect an AI agent to social media (MCP)
URL: https://skedcast.com/developers/mcp

Connect an AI agent to social media through SkedCast's remote MCP (Model Context Protocol) server, so it can read across your workspace (posts, clients, accounts, analytics, reports, best-times, library, and webhooks) and act on it — compose and edit posts, reschedule / cancel / retry deliveries, import media, and manage webhooks — with your permission, over a single OAuth-protected endpoint.

## Connect an AI agent to social media via MCP: the endpoint
Property | Value
Endpoint | `POST https://api.skedcast.com/mcp`
Transport | Streamable HTTP
Auth | OAuth 2.1 bearer token (audience-checked)
Discovery | `/.well-known/oauth-protected-resource`

## Add the server to your AI client
Point your MCP-capable client (Claude Desktop, IDEs, agents) at the endpoint below. The client runs the OAuth flow for you — it discovers the authorization server from the protected-resource metadata, registers itself (DCR), and opens a browser for you to sign in and consent. No manual app setup.

## Example MCP client config.
```json
{
  "mcpServers": {
    "skedcast": {
      "url": "https://api.skedcast.com/mcp"
    }
  }
}
```

## Available tools
Tool | Scope | Type | Description
`list_scheduled_posts` | `posts.read` | read | List the workspace's posts (queue, drafts, published) newest-first, with optional status / client / platform filters and free-text search (`q`) over captions, per-platform overrides, client names and account handles.
`list_post_targets` | `posts.read` | read | List individual post DESTINATIONS (one row per platform account) with optional status / client / account / platform / free-text filters and a SCHEDULE-time window — the tool for "what publishes on this date". Carries `total` and `totalPosts`, so a caller can tell a full page from a complete set.
`get_post` | `posts.read` | read | Get one post with its per-platform variants and fan-out targets.
`list_clients` | `clients.read` | read | List the workspace's client groups (brands / influencers).
`list_accounts` | `accounts.read` | read | List connected social accounts, optionally filtered by client or platform.
`get_analytics_overview` | `analytics.read` | read | Reach / impressions / engagement plus a per-platform rollup over a date window.
`get_post_analytics` | `analytics.read` | read | Per-post performance (complements get_analytics_overview): one post’s lifetime per-target breakdown by postId, or a keyset page of per-post rows with filters.
`get_account_analytics` | `analytics.read` | read | One channel’s totals and daily series, plus the two things no other tool carries: profile-level activity (profile views, link taps, video plays) and follower demographics.
`list_library_assets` | `media.read` | read | List content-library (DAM) assets with optional kind / folder / label / client / status / search filters. Filter `status: "transcoded"` for assets that are actually publishable; every row also carries `status`, `publishable` and `ready`.
`get_capabilities` | `posts.read` | read | The per-platform composing rulebook: caption and hashtag limits, media rules per kind (counts, formats, file-size ceiling, duration bounds, aspect ratios), post types, native-scheduling window, default cadence, first-comment support, and the `variants[].overrides` keys each platform requires or accepts. Read it before composing instead of discovering a limit by failing a publish.
`get_agency` | `agency.read` | read | The current workspace profile — name, plan, app mode, settings, and the caller’s role.
`list_reports` | `reports.read` | read | List the workspace's saved analytics report definitions.
`get_best_times` | `besttimes.read` | read | Worker-derived best posting times (strongest-first), optionally by client / platform.
`list_approvals` | `approvals.read` | read | List the workspace's approval queue (pending and decided requests), newest-first, with optional status / post filters.
`get_posting_limits` | `accounts.read` | read | Anti-ban posting limits (cadence): one account by socialAccountId, or every active channel plus the tenant bypass flag when omitted.
`get_quota` | `accounts.read` | read | Adaptive quota pool: your scarce-bucket budget (guaranteed slice, used, burst headroom, tomorrow’s guarantee tonight, reset clock) — read before planning a batch.
`list_campaigns` | `campaigns.read` | read | List the workspace's campaigns (named post groupings), optionally filtered by client / status.
`decide_approval` | `approvals.decide` | write ⚠ | Approve, reject, or request changes on a pending approval request (approve releases the gated post for publishing). A write action — confirm with the user before calling.
`validate_post` | `posts.compose` | read | Dry-run a compose request and get the per-platform verdict WITHOUT writing anything — same arguments as create_post, no post row and no draft. Returns `{ valid, platforms[], estimatedCostUsd }`. Does not run cadence or quota admission, so `valid: true` means the shape is acceptable, not that a publish cannot later be deferred.
`revise_post` | `posts.compose` | write ⚠ | Change the CONTENT of an already-scheduled post — caption, media, cover, alt text, per-platform overrides — without disturbing its destinations. Send only what you are changing. Refuses, naming them, when a destination has already published or is about to. An edit to an APPROVED post returns it to review. A write action — confirm with the user before calling.
`create_post` | `posts.compose` | write ⚠ | Compose a post and fan it out across the selected accounts (or save it as a draft). Retry-safe: retrying with identical arguments replays the original result instead of creating a duplicate. Note: two publish-now calls with byte-identical content and targets are treated as the same post and coalesced permanently — to intentionally publish the same content twice, vary the content or use distinct scheduled times. For platform-specific settings (e.g. YouTube privacy, TikTok privacy level, Pinterest board, Telegram chat), set `variants[].overrides`. That same object carries `firstComment`: a reply posted automatically once the destination goes live, so a "link in the comments" post needs no manual follow-up. Supported on X, LinkedIn, YouTube, Instagram, Facebook, Threads, Telegram and Bluesky (not TikTok or Pinterest); spell it exactly `firstComment`. Overrides are free-form — an unrecognised key is accepted and silently ignored — and the comment is best-effort: if it fails the post still publishes and nothing is reported, so confirm on the published post. A write action — confirm with the user before calling.
`create_posts_bulk` | `posts.compose` | write ⚠ | Create up to 50 posts in one call. Each item is exactly the arguments of create_post, so a bulk item carries every override a single post does (per-platform caption, title, cover, per-asset alt text, first comment, per-platform settings). Items are created one after another, each on its own: one failing never rolls back the others, and the result lists every item in order with its post id or its error. Retry-safe: retrying with identical arguments creates nothing twice. Pass validateOnly:true to check every item without writing anything. It counts one request-budget unit per item. This creates real outbound content — confirm with the user before calling. Requires the posts.compose scope.
`update_post` | `posts.compose` | write ⚠ | Replace a draft post in place — its content, media, targets, and schedule. Only draft posts can be updated, and this never publishes (use publish_draft for that). Platform-specific settings live in `variants[].overrides`, same as create_post, including `firstComment`. A write action — confirm with the user before calling.
`publish_draft` | `posts.compose` | write ⚠ | Publish or schedule an existing DRAFT post in place — same arguments as update_post plus the draft id; `schedule` decides now or later. Runs the same checks as create_post (platform validation, approval policy, quota, pacing); a client that requires approval gets the post submitted for review instead. The post keeps its id. This puts real content on connected accounts — confirm with the user before calling.
`import_media` | `media.manage` | write ⚠ | Ingest an external image / video into the content library by URL — the file must already be reachable at a public URL. For bytes you already hold (an image you just rendered, or a local file) use `upload_media` instead. A write action — confirm with the user before calling.
`upload_media` | `media.manage` | write ⚠ | Upload a LOCAL image / video into the content library and get back a media id for create_post. Sends the file inline (≤18 MiB) or returns a presigned URL for larger files. A write action — confirm with the user before calling.
`reschedule_post` | `posts.compose` | write ⚠ | Move a pending post target to a new time (absolute scheduledAt or relative shiftMinutes). A write action — confirm with the user before calling.
`bulk_post_target_action` | `posts.compose` | write ⚠ | Reschedule, retry or cancel up to 5000 post targets in one call — the batch form of reschedule_post / retry_post / cancel_post. Returns `{ jobId, accepted, skipped, skippedReasons? }`; a skipped target is reported, never silently dropped. A write action — confirm with the user before calling.
`cancel_post` | `posts.compose` | write ⚠ | Cancel a still-pending post target before it publishes. A write action — confirm with the user before calling.
`retry_post` | `posts.compose` | write ⚠ | Re-queue a failed or rate-limited post target so it publishes again. A write action — confirm with the user before calling.
`set_post_recycle` | `posts.compose` | write ⚠ | Set, update, or disable a post's evergreen recycle settings (interval / times / end date / max recycles). A write action — confirm with the user before calling.
`shift_campaign` | `campaigns.manage` | write ⚠ | Move a whole campaign's pending schedule by a relative number of minutes. A write action — confirm with the user before calling.
`list_webhooks` | `webhooks.read` | read | List the workspace's outbound webhook endpoints (url, subscribed events, active flag; secret masked).
`create_webhook` | `webhooks.manage` | write ⚠ | Register an outbound webhook endpoint for HMAC-signed event deliveries. Subscribe to any subset of the full 40-event catalog — see /developers/webhooks, or the `webhooks` block in openapi.json, for the complete list with payload fields. Returns the signing secret once. A write action — confirm with the user before calling.
`delete_webhook` | `webhooks.delete` | write ⚠ | Delete an outbound webhook endpoint by id (deliveries cascade). A write action — confirm with the user before calling.
`create_client` | `clients.manage` | write ⚠ | Create a new client group (a brand or influencer whose accounts are managed together). A write action — confirm with the user before calling.
`create_connect_invite` | `accounts.connect` | write ⚠ | Mint a shareable self-connect link so a client can connect the requested platforms themselves; returns the link and its expiry. A write action — confirm with the user before calling.
`get_webhook` | `webhooks.read` | read | Read one webhook endpoint — URL, subscribed events, active state, consecutive failures, and when/why it was auto-disabled. The signing secret is never returned here.
`list_webhook_deliveries` | `webhooks.read` | read | The delivery log for one endpoint — each attempt with its event type, response status, error and timing. How you find out WHY an endpoint stopped receiving.
`update_webhook` | `webhooks.manage` | write ⚠ | Change an endpoint’s URL, event list, or active state — including re-enabling one auto-disabled after repeated failures. The event list REPLACES the previous one. A write action — confirm with the user before calling.
`test_webhook` | `webhooks.manage` | write ⚠ | Send a `webhook.test` delivery and report the outcome, proving an endpoint is reachable and verifying signatures without waiting for a real event. Performs a real outbound request — confirm with the user before calling.
`rotate_webhook_secret` | `webhooks.manage` | write ⚠ | Mint a new signing secret and return it ONCE — it is not readable again. The previous secret keeps verifying for a short overlap so the customer can redeploy without dropping deliveries. A write action — confirm with the user before calling.
`get_campaign` | `campaigns.read` | read | Read one campaign in full — name, description, status, colour, date range, owning client and post counts.
`create_campaign` | `campaigns.manage` | write ⚠ | Create a campaign — a named group of posts with an optional client, colour and date range. Posts are added with assign_to_campaign. A write action — confirm with the user before calling.
`update_campaign` | `campaigns.manage` | write ⚠ | Change a campaign’s name, description, status, colour or date range. A write action — confirm with the user before calling.
`delete_campaign` | `campaigns.manage` | write ⚠ | Delete a campaign. The posts in it are NOT deleted — they lose the grouping and keep their own schedules. A write action — confirm with the user before calling.
`assign_to_campaign` | `campaigns.manage` | write ⚠ | Add up to 500 existing posts to a campaign. Grouping only — no post’s content or schedule changes. A write action — confirm with the user before calling.
`unassign_from_campaign` | `campaigns.manage` | write ⚠ | Remove posts from a campaign. Grouping only — the posts and their schedules are untouched. A write action — confirm with the user before calling.
`get_client` | `clients.read` | read | Read one client group in full — name, type, approval policy, posting rules, brand voice and default timezone.
`update_client` | `clients.manage` | write ⚠ | Change a client’s name, type, approval policy, posting rules, brand voice or default timezone. A write action — confirm with the user before calling.
`delete_client` | `clients.manage` | write ⚠ | Archive a client group. OWNER ONLY — a Manager holding the clients.manage scope cannot do this. A write action — confirm with the user before calling.
`get_post_target` | `posts.read` | read | Read a single fan-out destination in full — status, scheduled time, platform, account, published permalink and external id, and its failure reason when it failed.
`get_post_recycle` | `posts.read` | read | Read one post’s evergreen re-posting configuration and the worker’s bookkeeping. Returns null when the post is not enrolled.
`get_account` | `accounts.read` | read | Read one connected social account in full — platform, handle, display name, avatar, status (including whether it needs re-authorisation) and owning client.
`get_approval` | `approvals.read` | read | Read one approval request with the gated post’s preview and roll-up, so a reviewer can see what they are deciding on before calling decide_approval.
`get_media_asset` | `media.read` | read | Read one content-library asset in full — filename, kind, size, dimensions, duration, folder, labels, alt text, pipeline status, and the publishable/ready flags plus the per-platform readiness map.
`get_client_analytics` | `analytics.read` | read | One client group’s analytics — totals, a per-platform breakdown and a time series. A different projection from get_analytics_overview filtered by client.
`get_report` | `reports.read` | read | Read one report definition with its run history — each run’s status, window, and whether an export is available.
`download_report_export` | `reports.read` | read | Get a time-limited download URL for one completed report run’s export file. The URL expires — fetch it when you are ready to use it.
`clear_post_recycle` | `posts.compose` | write ⚠ | Turn OFF evergreen re-posting for a post. Idempotent. Already-scheduled recycled copies are not retracted — cancel those separately. A write action — confirm with the user before calling.

## Write tools confirm first
The write tools — create_post, update_post, import_media, reschedule_post, cancel_post, retry_post, create_webhook, and delete_webhook — change your workspace. They're marked non-read-only, so a well-behaved host prompts you to confirm before the agent calls them. As always, the agent's effective power is your scopes ∩ your role.

## Discovery URLs
URL | Purpose
`https://api.skedcast.com/mcp` | MCP endpoint (POST; GET returns 405)
`/.well-known/oauth-protected-resource` | Resource metadata + scopes (RFC 9728)
`/.well-known/oauth-authorization-server` | Authorization-server metadata (RFC 8414)
`/.well-known/jwks.json` | Public keys for token verification

## FAQ
**Q: Which AI clients can connect?**
A: Any client that speaks the Model Context Protocol over Streamable HTTP with OAuth — including Claude and a growing list of agent frameworks and IDEs.

**Q: Can the agent post without my approval?**
A: Read tools run freely within your granted scopes. The write tools (create_post, import_media) are flagged so your MCP host asks you to confirm before they run, and they still can't exceed your role.