Documentation
Parker MCP Set Up
Connect Parker to Claude, Cursor, Windsurf, Claude Code, and other MCP clients using OAuth or an API token.
Getting Started
Step 1 — Create a Parker Account
Sign up for a Parker account at heyparker.ai. You'll need an active paid subscription to use the Parker connector.
Step 2 — Set Up Your Brand Workspace
After logging in to Parker, make sure your brand workspace is set up, your Meta ad account is connected, and that you completed onboarding to make sure Parker is trained on your account info. Parker uses this workspace as the source for all ad data, reviews, and creative research.
FYI- it usually takes 2 - 6 hours (can be up to 24 hours if very large) for your Meta account to be fully synced and analyzed with Parker
Step 3 — Get Your MCP Server URL
All you need to connect to Claude is the Parker MCP server URL:
https://api.heyparker.ai/api/mcp/full
Copy it and continue to Step 4a to connect with OAuth (recommended for Claude). You don't need to visit any settings page if you already have this URL.
Need an API token, or want to copy the URL from the app? In Parker, open the account menu and go to Settings → MCP. Under Parker MCP you can copy the same URL and choose between two connection methods:
- OAuth (recommended for Claude and other interactive clients) — browser sign-in, no token to manage. Follow Steps 4a–6.
- API token (Config file + Bearer) (best for Cursor, Windsurf, Claude Code, and scripts) — MCP-spec compliant; the token stays in your client's config headers. You can also generate or rotate your token here. Follow Step 4b.
Step 4a — Connect with OAuth (Claude)
On the OAuth tab, copy the MCP server URL (https://api.heyparker.ai/api/mcp/full). No API token or brand ID is required.
Then in Claude.ai, navigate to Settings → Connectors and choose Add custom connector:

- Name: Parker
- URL:
https://api.heyparker.ai/api/mcp/full

Click Add.
Note: Parker MCP uses the Streamable HTTP transport protocol. Adding Parker as a custom connector (rather than from the Claude Directory) lets you rename it if desired. The same steps apply across Claude chat (web and desktop), Claude Cowork, and Claude Code.
Step 5 — Connect and Authorize
Go to your Claude Settings → Connectors, find Parker in your list, and click Connect. You'll be taken through an authorization flow — sign in on the Parker login page, then click Allow access on the consent screen to complete the connection.
Step 6 — Start Querying
As soon as Parker is connected, you can query against any of the brands you have access to in Parker. On your first message, Claude will pull your available brands so you can pick which one to work with, and you can switch between them anytime.
Step 4b — Connect with an API Token (Cursor, Windsurf, Claude Code, scripts)
On the API token tab, first Generate your API token (create a personal access token, or copy the existing one) — your MCP client sends it as a Bearer token on every request. Then, under Choose how to connect, pick the format your client expects.
Option A — Config file (recommended) — for Cursor, Claude Code, Windsurf, and most coding agents. Copy the ready-made MCP config JSONand paste it into your client's config file. It uses the Authorization header — the MCP-spec-compliant way to authenticate:
{
"mcpServers": {
"hey-parker-full": {
"transport": "streamable-http",
"url": "https://api.heyparker.ai/api/mcp/full",
"headers": {
"Authorization": "Bearer YOUR_PARKER_API_TOKEN"
}
}
}
}Replace YOUR_PARKER_API_TOKEN with your token. Transport is Streamable HTTP; auth is Bearer token. Common config file locations:
- Cursor:
~/.cursor/mcp.json - Claude Code:
~/.claude.json - Project scope:
.mcp.jsonin your project root
Save and reload your client — Parker authenticates automatically via the Bearer token. If your client isn't listed, look for "MCP" or "Model Context Protocol" in its settings.
Option B — Direct URL— for pasting into Claude's custom connector when the OAuth redirect fails. Copy the Direct connect URL (your token is already embedded):
https://api.heyparker.ai/api/mcp/full?access_token=YOUR_PARKER_API_TOKEN
In Claude, go to Settings → Connectors → Add custom connector, paste the Direct URL, name it Parker, and click Add. Same flow as OAuth, but the token is already in the URL — no redirect.
Keep your token secret.Don't paste your real API token or Direct connect URL into shared docs, repos, or screenshots. If a token is exposed, use Rotate token on the MCP page — note that rotating it will break any connected clients until you update their config with the new token.
After your client connects, your first message will prompt you to select a brand, just like the OAuth flow.
Available Tools
Tools you can use through Parker MCP, with short descriptions and annotations (read-only vs write, destructive, and so on).
Tools below can appear when you connect to Parker MCP. Availability can vary by account and plan.
Brand scoping: Parker MCP does not pre-select a brand. Call get_available_brands first, have the user pick a brand, then pass brand_id / brandId on every brand-scoped tool call.
Annotation meanings
- read-only — intended not to change your data
- write — creates or updates data
- destructive — can delete or irreversibly change data
- idempotent — safe to retry with the same inputs
- open-world — may reach the public web or systems outside Parker
Brand discovery
Get available brands
get_available_brands
MCP-onlyLists organizations and brands the authenticated user can access. Call this first. After the user picks a brand, pass that id as brand_id / brandId on subsequent brand-scoped tools.
read-onlyidempotent
Brand context & memory
Get brand persona
get_brand_persona
Loads the brand persona / brand context for the active brand.
read-onlyidempotentUpdate brand context
update_brand_context
Updates brand context from a plain-English instruction (add, change, or remove content). This is a write-only tool — it starts the update and returns right away; the change finishes in the background.
writeUpdate working memory
update_custom_working_memory
Creates, updates, or deletes long-term org / brand / user memories that persist across conversations.
writedestructive
Facebook ads & attribution
Search Facebook ads (SQL)
search_facebook_ads_sql
Primary structured search over the brand's Facebook ads performance (metrics, filters, and date ranges).
read-onlyidempotentSearch Facebook ads (semantic)
search_facebook_ads_semantic
Semantic search over indexed Facebook ad reports / creatives.
read-onlyidempotentList custom metrics
list_custom_metrics
Lists the brand's custom conversion / event / metric and formula definitions.
read-onlyidempotentSearch Northbeam attributed ads
search_northbeam_attribution
Returns Northbeam-attributed performance for the brand's Facebook ads when Northbeam is connected; otherwise falls back to Facebook performance data.
read-onlyidempotentCheck Northbeam connection
check_northbeam_connection
Returns whether Northbeam is connected for the brand (connected: true|false) and the metrics preference when connected. Call before search_northbeam_attribution when unsure; when false, use Meta ads tools and do not mention Northbeam unless the user asked.
read-onlyidempotentSearch Facebook ad comments (SQL)
search_facebook_ad_comments_sql
Structured search over Facebook ad comments.
read-onlyidempotentSearch Facebook ad comments (semantic)
search_facebook_ad_comments_semantic
Semantic search over Facebook ad comments.
read-onlyidempotent
Customer reviews & surveys
Search customer reviews (SQL)
search_customer_reviews_sql
Structured filter and search over customer reviews.
read-onlyidempotentSearch customer reviews (semantic)
search_customer_reviews_semantic
Semantic search over customer reviews.
read-onlyidempotentSearch post-purchase surveys (semantic)
semantic_search_post_purchase_survey
Semantic search over post-purchase survey responses.
read-onlyidempotentLookup post-purchase surveys (numeric)
lookup_post_purchase_survey
Numeric / filtered lookup of post-purchase survey data.
read-onlyidempotent
Competitors, organic & websites
Brand discovery
brand_discovery
Corpus matches/compare; one-off ingest for a named missing brand; batch discover_net_new (web research → up to ~10 competitors, confirm:true, 12h cooldown) + discover_net_new_status. Write path — not read-only. Feature flag: brand_discovery.
writeopen-worldSearch competitor Facebook ads
search_competitor_facebook_ads
Searches and manages ads from the Meta Ad Library for competitor and inspiration brands.
writeSearch and manage organic social
search_and_manage_organic_social
Search and manage organic social posts across supported platforms.
writeMonitor brand websites
monitor_brand_websites
Brand website intelligence: positioning, audience, products, promos, page changes, and comparisons.
writeopen-world
Search Reddit posts and comments
search_reddit_posts_and_comments
Searches Reddit posts and comments tracked for the brand (structured and semantic lookup).
read-onlyidempotent
Video
Search TikTok videos
search_tiktok_videos
Searches Parker's TikTok and short-form video index.
read-onlyidempotentAnalyze video from URL
analyze_video_from_url
Analyzes a public video URL for creative insights.
writeopen-world
Web & utilities
Summarize webpage
get_webpage
Fetches and summarizes text content for a given URL.
read-onlyopen-worldSearch the web
search_web
Live web search for current information.
read-onlyopen-worldGet current time
get_current_time
Returns the current date and time for the request.
read-onlySearch chat history
search_chat_history
Searches past Parker chat threads and messages accessible to the user.
read-onlyidempotentManage insight subscriptions
manage_insights_subscriptions
Creates, updates, or deletes insight subscription schedules and preferences.
writedestructiveopen-world
Parker Brain
Update Parker Brain setup status
update_parker_brain_setup_status
MCP-onlyRecords setup progress (phases and completion) for a brand's Parker Brain.
writeSet up Parker Brain storage
setup_parker_brain
MCP-onlyProvisions or refreshes Parker Brain storage and access for a brand. Does not generate brain content itself — the calling agent writes into that storage afterward.
writedestructiveopen-world
Troubleshooting
Parker isn't responding or the MCP connection fails
- Confirm your Parker subscription is active at heyparker.ai
- Verify the URL is
https://api.heyparker.ai/api/mcp/full(copied from Settings → MCP in Parker) - If you're using an API token, confirm the
Authorization: Bearer …header is set correctly and the token hasn't been rotated - If while using OAuth an issue occurs, try using the Direct URL instead of the general URL above.
- Try disconnecting and reconnecting the Parker connector in Claude Settings → Connectors
- If re-connecting via OAuth, complete the authorization flow and click Allow access
- If the issue persists, contact support at support@realsimplelabs.com or use your #parker slack channel
Parker responds but doesn't have my brand's data
- Make sure you selected the correct brand when prompted — ask Claude to show your available brands again if you need to switch
- Confirm your Meta ad account is connected inside Parker
for newly onboarded brands, it usually takes 2 - 6 hours (can be up to 24 hours if very large) for your Meta account to be fully synced and analyzed with Parker
I'm getting an authentication error
- OAuth: disconnect and reconnect the Parker connector in Claude Settings → Connectors to re-trigger the OAuth flow
- Use the Direct URL under the "API token" page instead of the general URL
- API token: make sure the token in your config matches the current token on the MCP page (rotating the token invalidates the old one)
- Make sure you're using the same Parker account that has an active subscription
- If the issue persists, contact support at support@realsimplelabs.com or use your #parker slack channel
Support
For help with Parker MCP, contact us at support@realsimplelabs.com
If you're an active Parker customer, you also have access to a dedicated Slack channel created during your onboarding — that's the fastest way to get support.
Privacy & Data
Parker's privacy policy is available at: https://heyparker.ai/support/privacy-policy
Parker transmits all data over HTTPS/TLS. Parker only accesses data explicitly requested by the user during a session.

