Agent access · MCP

Connect your AI agent to ChannelFuel

ChannelFuel is an MCP server: every step of the journey — topic scans, video breakdowns, scripts, titles and thumbnails — is a tool your own AI agent can call. Agent calls spend the same credits and face the same checks and grades as the buttons in the app, so an agent can't produce slop through ChannelFuel any more than a human can.

Endpoint · streamable HTTP

https://channelfuel.app/api/mcp

Authentication

Sign in with your browser — add the URL, click Approve, done.

Standard MCP OAuth: your client opens this site, you approve once. Prefer a key for scripts? Create one in Account → Agent access. Revoke either anytime.

Set up your client

Add the endpoint with no credentials — your client opens the browser and you click Approve. Every block below works that way; the API-key variant is at the end for scripts and CI.

claude.ai + Claude Desktop

Settings → Connectors → Add custom connector → paste the URL. Claude opens the sign-in page; approve and the tools appear in every chat.

https://channelfuel.app/api/mcp

Claude Code

One command, then run /mcp inside a session to trigger the browser sign-in.

claude mcp add --transport http channelfuel https://channelfuel.app/api/mcp

Cursor

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global). Cursor shows 'Needs login' next to the server — click it and approve in the browser.

{
  "mcpServers": {
    "channelfuel": { "url": "https://channelfuel.app/api/mcp" }
  }
}

OpenAI Codex CLI

Add to ~/.codex/config.toml, then run `codex mcp login channelfuel` to open the browser sign-in.

[mcp_servers.channelfuel]
url = "https://channelfuel.app/api/mcp"

VS Code (Copilot agent mode)

Add to .vscode/mcp.json and start the server from the MCP view — VS Code walks you through the browser sign-in.

{
  "servers": {
    "channelfuel": { "type": "http", "url": "https://channelfuel.app/api/mcp" }
  }
}

Anything else (generic bridge)

Any stdio-only client can bridge with mcp-remote — it handles the browser sign-in itself:

npx -y mcp-remote https://channelfuel.app/api/mcp

Scripts, CI & the Claude API — use an API key

Headless callers can't click Approve. Create a key in Settings → Agent access and send it as a bearer token; on the Claude Messages API pass it as authorization_token.

# any HTTP client
Authorization: Bearer cf_live_…

# Claude API
"mcp_servers": [{
  "type": "url",
  "url": "https://channelfuel.app/api/mcp",
  "name": "channelfuel",
  "authorization_token": "cf_live_…"
}]

Under the hood: the endpoint implements standard MCP OAuth — discovery via /.well-known, dynamic client registration, PKCE, refresh-token rotation. Any client that follows the MCP auth spec connects with zero manual credentials, and every connection shows up (and can be revoked) in Settings → Agent access.

The tools

Credited tools debit your workspace balance exactly like the UI — and refund automatically if the job fails. The checks are enforced server-side: no script without a broken-down proven video, no thumbnail before a chosen title, and glance-test overrides are logged against your own results. (Tool names are stable API identifiers — in the app these are video breakdowns, topics, story plans and the glance test.)

get_workspacefree

Workspace, plan, credit balance + price list

list_channels / list_stylesfree

Channels with locked look + format; the style catalog

list_projects / get_projectfree

Where every video sits in the six steps — script, packaging, publishes

list_teardowns / get_teardownfree

Your video breakdowns; one full breakdown of a proven video

get_niche_reportfree

A finished topic scan with evidence grades

run_demand_sweep25 cr

Scan seed topics for small-channel wins, grouped into topics

get_demand_familyfree

One topic's evidence: proof it repeats, competition, member videos

list_ideas / create_ideafree

Your ideas — evidence from the topic finder + your twist, four grades each

grade_novelty10 cr

Grade an idea's twist against its topic — clashing titles quoted

run_niche_scan25 cr

Scan YouTube for small-channel wins on a topic, graded on evidence

run_teardown40 cr

Break down a proven video — how it earned its views

create_video_projectfree

Start a video on a channel (long-form or Short)

prefill_worksheet10 cr ×model

Draft the story plan from the broken-down proven video

generate_script150 cr ×model ×length

Script written scene by scene + scorecard

generate_titles15 cr ×model

6-8 glance-tested title + hook concepts

generate_thumbnails60 cr

3 thumbnail variants for the chosen title

generate_metadata10 cr ×model

Spam-safe description + tags

choose_packagingfree

Pick title/thumbnail — glance test applied, overrides logged

Publishing to YouTube isn't exposed over MCP yet — connect the channel and schedule from the app. Agents also can't bypass plan limits: publishing needs a paid plan either way.

Ready to hand your agent the keys?

Add the endpoint, click Approve in the browser, and your agent is in. New workspaces start with 200 free credits — enough for a topic scan, a video breakdown and a script.

Start free