Agent Quickstart
The fastest path from an MCP-aware agent to a first governed call: endpoint, authentication, discovery, first call.
For AI agents, and the humans configuring them. Everything you need to connect to Mediafier, in one short page. The rest of this documentation is depth; none of it is required to make your first call.
What Mediafier is
Mediafier connects enterprise media systems to AI agents through a secure, audited, commercially governed layer. You get tools (media processing, library, metadata, intelligence), governed resources, and MediaClaws — autonomous multi-step workflows — through one standard MCP endpoint. The operating principle: agents think, gateways decide, tools act.
Access state
Invite-only pre-launch. This documentation is public; executing anything requires an invited account and organization. No access? Hand mediafier.ai/request-access to your human and stop — retrying authentication will not change the outcome. Details: Getting Access.
The endpoint
POST https://mcp.mediafier.ai/mcpMCP over Streamable HTTP, JSON-RPC 2.0. This is the only canonical ingress — there is no per-capability base URL to discover and no tenant header to set. Interactive harnesses sign in with OAuth; headless and automation flows exchange a Mediafier-issued Agent Credential for an access token (see Authentication & Agent Credentials).
Connect your client
Setting up the client you are running in? Follow mcp.mediafier.ai/connect: find your client, add the server, sign in, verify.
How authentication works
Sign-in is OAuth through the endpoint (harnesses that support MCP auth handle it natively — in Claude Code, /mcp starts it). Your organization is derived server-side from your identity; a client never selects a tenant. Unauthenticated initialize succeeds with capability negotiation; protected calls return 401 with standard WWW-Authenticate metadata. A 401 can mean several things: not signed in, an expired session, or an account with no invited organization. If sign-in completes and calls still return 401, the likely cause is missing access, not a broken client — see Getting Access.
What you can discover
Discovery is JSON-RPC, not a documentation page:
tools/list— the tools your authorization actually grants (sign in first; discovery is authorization-scoped)resources/list— governed resources visible to your organization
Take tool names, argument shapes, and bounds from these live responses — they are the operational source of truth for what your authorization grants right now.
What the words mean
Governed resource — anything invocable (a tool, a skill, a MediaClaw, a workflow) registered with the platform and subject to the same authorization, billing, and audit chain.
MediaClaw — a governed autonomous workflow bonded to one outcome (for example: ingest a podcast → transcribe → chapter → brand-safety-check → publish metadata). Long-running; you start one, then poll its status.
Decision trace — the inspectable record of what an agent did and why. Every call you make is traced; your humans can audit it.
Calls cost credits
Execution is metered in credits held by your organization. When a call would exceed the balance you get a structured 402 naming the exact shortfall and a billing URL for your human — top-ups are a human action, never yours.
Where to go next
Access state, refusal diagnosis
Setup for a specific client (Claude, Codex, Cursor, SDKs)
How enforcement works
Autonomous workflows
Machine-readable version of this orientation: mcp.mediafier.ai/llm.txt.
Last updated