MCP integration guide

Connect any MCP-compatible AI agent to AutoPersonas — no SDK, no HTTP plumbing. Paste one URL into your client and the agent can manage characters, generate content batches, approve drafts, and publish to social. The hosted MCP server at https://mcp.autopersonas.com exposes 140 tools, 7 resources, and 6 prompts that wrap our developer REST API.

Quickstart

Pick your client

Click the entry for the agent you're setting up. Each one accepts the same URL — the OAuth flow handles the rest.

// OAuth — recommended
{
  "mcpServers": {
    "autopersonas": {
      "url": "https://mcp.autopersonas.com"
    }
  }
}

// API key — for terminal-only setups
{
  "mcpServers": {
    "autopersonas": {
      "url": "https://mcp.autopersonas.com",
      "headers": { "Authorization": "Bearer ap_live_..." }
    }
  }
}

# OAuth (recommended)
claude mcp add --transport http autopersonas https://mcp.autopersonas.com

# API key
claude mcp add --transport http autopersonas https://mcp.autopersonas.com \
  --header "Authorization: Bearer ap_live_..."

Authentication

AutoPersonas's MCP supports two credential types — pick the one that matches your client:

Test the connection

Ask your agent "list my AutoPersonas tools". You should see the autopersonas_* set, plus the /autopersonas: slash-command prompts. Run autopersonas_account_info as a smoke test — it returns your user ID, scopes, and remaining rate-limit headroom.

What you can do

The MCP server surfaces the full developer-API as 140 tools spread across the categories below. Most map one-to-one to a REST endpoint; a few wrap multi-step flows (poll-until-complete, batch-and-wait) for agent ergonomics.

Costs are sourced from src/lib/api-pricing.ts at runtime — figures shown here are typical published rates and may drift slightly with provider pricing changes.

Slash commands

Prompts are reusable orchestration templates that collapse a multi-step flow into a single command. They're invoked from the agent (Claude, Cursor, etc.) with the /autopersonas: prefix.

Scopes

Each tool requires a scope; the "Connect AI agent" button defaults to read + generate (no *:write for safety). Grant additional scopes from the API Keys page or mint a separate key per agent.

Cost

Tools that incur charges display the cost in their description so the agent can surface it in chat before invoking. Examples (in cents, USD):

• autopersonas_generate_content — ~20¢–36¢ / image (depending on resolution + provider).
• autopersonas_publish_content — 2¢ / post + per-platform fees.
• autopersonas_generate_character_references — ~80¢ for a 4-view pack.

Live rates come from src/lib/api-pricing.ts. Your monthly budget cap (configured under Billing) applies to MCP-driven calls just like dashboard-driven calls.

Examples

Create a branded persona end-to-end

/autopersonas:create-branded-persona \
  name="Maya" \
  niche="beauty" \
  brandGoals="EarthGlow Serum, RecycleHair Brush"

The agent will: create the character, attach brand goals, confirm. Use autopersonas_get_character afterward to inspect the result.

Generate a 2-week content batch

/autopersonas:batch-generate-and-schedule \
  profileId="prof_abc123" \
  characterId="char_xyz789" \
  startDate="2026-05-01" \
  endDate="2026-05-14" \
  postsPerWeek="5"

Each post is generated asynchronously; the prompt waits on the batch and returns the resulting drafts so you can review and approve them in one shot.

Publish without duplicates

/autopersonas:publish-with-dedup-check \
  contentId="ct_42" \
  platforms="x,instagram"

Troubleshooting

Click an error to see what causes it and how to fix it.

Limitations

• OAuth-aware clients only: Claude Desktop, Claude.ai web, Cursor, Cline, Continue, VS Code Copilot agent mode all support the OAuth flow. Older or scripted MCP clients may not — use the API-key path instead.
• Rate limit: 60 requests/minute per user across all keys.
• Long-running ops: content batches return a batchId and have a companion autopersonas_wait_for_batch tool. LoRA training jobs use autopersonas_wait_for_lora for the same pattern.

Privacy

The bearer key in your agent's config syncs via the OS keychain (Claude Desktop) or wherever your terminal / IDE extension stores it (Claude Code, Cursor, Cline, Continue, …). We strongly recommend minting a separate MCP-scoped key (the Connect AI agent default) so a token leak doesn't compromise your full-scope developer key — or use OAuth, which never puts a long-lived secret on disk in the first place. Revoke any leaked key immediately from the API Keys page — disabled keys reject with 401 on the next request.

Related

• Developer API — the REST surface the MCP server wraps.
• modelcontextprotocol.io — the open standard.