The hosted MCP server

One Streamable-HTTP endpoint that gives any Model Context Protocol client — Claude, Claude Code, Cursor, Codex — six tools for running real marketing work.

MCP endpoint

https://growcita.com/api/mcp

Get an API key

Authentication

Two ways to present your key — use the header whenever your client supports it:

Preferred — Authorization header

Authorization: Bearer gcta_YOUR_KEY

Fallback — query parameter (clients that can't set headers)

https://growcita.com/api/mcp?key=gcta_YOUR_KEY

URLs can end up in proxy and gateway logs, so the ?key= form exists only for clients without header support (claude.ai custom connectors). If a key may have leaked, revoke it in Settings → API Keys — revocation is immediate.

Connect from claude.ai, Claude Desktop & Cowork

1Open Settings → Connectors (claude.ai, Claude Desktop, or Cowork).
2Choose "Add custom connector".
3Paste the endpoint with your key: https://growcita.com/api/mcp?key=gcta_YOUR_KEY
4Save, then enable the connector in a conversation and ask Claude to list your Growcita sessions.

Working in Claude Code, Cursor, or Codex instead? Those clients support the Authorization header — follow the dedicated guides: Claude Code, Cursor, Codex.

The six tools

Parameters marked * are required. Full JSON Schemas are returned by tools/list.

list_agents

no parameters

List Growcita's 10 marketing agents with their specialties and sub-specialties. Use this to decide which agentType to pass to run_agent_task. The growth-marketer agent is the supervisor: give it any free-form marketing objective and it routes work to the right specialist itself.

list_sessions

limit

List the user's Growcita workspaces (sessions). Every task runs inside a session, which carries the company context (name, website, industry, audience, goals) that grounds agent output. Call this first to get a sessionId for run_agent_task. Most users have one session per company they market.

create_session

name*, companyName, companyUrl, industry, targetAudience

Create a new Growcita workspace (session) for a company. Only needed when list_sessions shows no suitable workspace — prefer reusing an existing session because it already carries learned company context, connected platforms, and history. Note: sessions created via the API skip the automatic onboarding research crawl, so include companyName and companyUrl for best results.

run_agent_task

sessionId*, instruction*, agentType, title, priority, waitForCompletionSeconds

Run a marketing task with one of Growcita's agents. Pass the instruction in plain English (e.g. "Audit our landing page for conversion blockers" or "Draft a week of LinkedIn posts about our new feature"). Tasks run asynchronously in the background and typically take 1–5 minutes; this tool returns the task with its current status. Set waitForCompletionSeconds (max 120) to poll server-side before returning, and use get_task to poll afterwards if the status is still pending or running. Actions that publish content or change campaigns are guarded: they may finish as a draft or wait for human approval in the Growcita app rather than going live directly.

get_task

taskId*

Fetch a task's current status and result. status is one of pending, running, completed, failed. The result field contains the agent's full output once completed; error explains failures. Poll this (suggested interval: 10–30s) after run_agent_task until the task reaches a terminal state.

list_tasks

sessionId, status, agentType, limit

List recent tasks, optionally filtered by sessionId, status, or agentType. Use it to check what's running, find a lost taskId, or review recently completed work.

What a session looks like

Prompts you can paste into a connected client:

prompt

List my Growcita sessions, then ask the SEO agent to audit our site for technical issues and report the top 10 fixes.

prompt

Use run_agent_task with the growth-marketer agent: plan next week's marketing priorities based on our goals. Wait for completion, then summarize.

prompt

Check the status of my running Growcita tasks and show me the result of the most recently completed one.

Tasks run asynchronously and typically take 1–5 minutes — clients poll with get_task. Actions that publish or change campaigns follow your autonomy and approval settings, so a result may say a draft is waiting for review in the app. That's by design.

stdio wrapper: @growcita/mcp

Some clients prefer launching a local process over connecting to a URL. The npm package is a thin stdio proxy to this endpoint — same six tools, key supplied via environment variable.

bash

GROWCITA_API_KEY=gcta_YOUR_KEY npx -y @growcita/mcp