trakkr-mcp

trakkr-mcp

MCP server for Trakkr — AI visibility monitoring.

Connect your AI assistant (Claude Code, Claude Desktop, Cursor, VS Code, Codex CLI, Windsurf, Zed, Cline, Continue, and any other MCP client) to Trakkr and query your brand's AI search visibility data conversationally.

Quick Start

The fastest path is uv — uvx auto-installs and runs the server with zero setup:

# One-time, if you don't have uv:
brew install uv

# Set your MCP connect token from Settings > Developer:
export TRAKKR_API_KEY="mcp_connect_your_token_here"

# Optional, run the server directly to verify it works:
uvx trakkr-mcp

You usually don't run uvx trakkr-mcp yourself — your MCP client runs it for you based on the config below. Scale users with legacy sk_live_ REST keys can still use those keys for direct local mode, but new setups should use the MCP connect token.

Official MCP Registry

Registry name: io.github.macklpgr/trakkr-mcp

The Registry entry offers both supported connection paths:

• Hosted remote: https://api.trakkr.ai/mcp over Streamable HTTP. Compatible clients discover Trakkr's OAuth endpoints and ask the user to authorize the connection.
• Local package: uvx trakkr-mcp over stdio. Set TRAKKR_API_KEY to an MCP connect token, or omit it to use the device authorization flow.

The source repository is currently private. The published PyPI artifacts, this package README, the hosted OAuth metadata, and the public product documentation are the reviewable release surfaces. The Registry metadata does not claim that the private repository is publicly browsable.

Configuration

Pick the client you use. The full setup guide, including a live picker that updates the snippet per client, lives at trakkr.ai/learn/api/mcp.

Claude Code (CLI)

Run once in any terminal:

claude mcp add trakkr \
  -e TRAKKR_API_KEY=mcp_connect_your_token_here \
  -- uvx trakkr-mcp

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "trakkr": {
      "command": "uvx",
      "args": ["trakkr-mcp"],
      "env": {
        "TRAKKR_API_KEY": "mcp_connect_your_token_here"
      }
    }
  }
}

Cursor

~/.cursor/mcp.json (global) or .cursor/mcp.json (project):

{
  "mcpServers": {
    "trakkr": {
      "command": "uvx",
      "args": ["trakkr-mcp"],
      "env": {
        "TRAKKR_API_KEY": "mcp_connect_your_token_here"
      }
    }
  }
}

VS Code (Copilot Chat)

.vscode/mcp.json — uses the input-prompt pattern so the key never lands in source:

{
  "servers": {
    "trakkr": {
      "type": "stdio",
      "command": "uvx",
      "args": ["trakkr-mcp"],
      "env": {
        "TRAKKR_API_KEY": "${input:trakkr-connect-token}"
      }
    }
  },
  "inputs": [
    {
      "type": "promptString",
      "id": "trakkr-connect-token",
      "description": "Trakkr MCP connect token",
      "password": true
    }
  ]
}

Codex CLI

~/.codex/config.toml:

[mcp_servers.trakkr]
command = "uvx"
args = ["trakkr-mcp"]
env = { TRAKKR_API_KEY = "mcp_connect_your_token_here" }

Windsurf

~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "trakkr": {
      "command": "uvx",
      "args": ["trakkr-mcp"],
      "env": {
        "TRAKKR_API_KEY": "mcp_connect_your_token_here"
      }
    }
  }
}

Zed

~/.config/zed/settings.json:

{
  "context_servers": {
    "trakkr": {
      "source": "custom",
      "command": {
        "path": "uvx",
        "args": ["trakkr-mcp"],
        "env": {
          "TRAKKR_API_KEY": "mcp_connect_your_token_here"
        }
      }
    }
  }
}

Cline, Continue, or other MCP clients

Most clients accept the standard mcpServers JSON shown above for Cursor/Claude Desktop/Windsurf. Check your client's MCP documentation for the exact config file location, and see trakkr.ai/learn/api/mcp for an up-to-date picker.

Available Tools

The heaviest read tools (get_citations, get_competitors, get_reports, get_research_run) take a response_format argument: concise (the default) caps long lists and drops verbose dumps so the result never swamps the model's context, while detailed returns the full payload. Either way a size guard holds every response under the context ceiling and, when it trims, tells you how to fetch the rest (page with offset, narrow the window, or read the matching trakkr:// resource).

Core Data

Tool	Description
report_unmet_need	Record a request that no current Trakkr tool can complete. This is a product signal only and does not perform the missing action.
list_brands	List all brands you're tracking. Returns brand IDs needed for other tools.
get_visibility_scores	Get AI visibility scores and trends over time.
list_prompts	List tracked search queries for a brand.
list_tags	List all prompt tags for a brand, including unused tags and current prompt counts.
manage_prompt	Create, update, or delete a tracked prompt.
manage_prompt_tags	Create, rename, recolour, or delete a tag, or add/remove it across up to 200 prompts. Add and remove are safe to retry.
suggest_prompts	Suggest fresh prompts to track, grounded in the brand's own profile and deduped against what it already tracks.
update_brand_aliases	Update the brand's aliases (alternate names folded into its visibility). Add, remove, or set.

Visibility

Tool	Description
get_citations	Citation URLs and trends. Views: list, history, queries, sources, feed, heatmap.
get_rankings	Competitive rankings in AI search results.
get_model_breakdown	Visibility by AI model (ChatGPT, Perplexity, Gemini, etc.).
get_competitors	Competitor analysis. Views: summary, arena, head-to-head, by-model, threats.
manage_competitor	Track, untrack, hide, or unhide a competitor. add/remove touch the tracked set; hide/unhide suppress a competitor from views; list reads current state.

Intelligence

Tool	Description
get_opportunities	Citation gaps where competitors appear but you don't.
get_content_ideas	AI-generated content ideas to improve visibility.
get_perception	How AI models describe and position the brand. Paid plan required.
get_prism	AI Pages analysis: positioning, strengths, weaknesses, opportunities. Paid plan required.
get_narratives	Narrative intelligence: tracked topics and storylines. Scale plan required.

Research

Read access to the full prompt research analytics — the same payload that powers the in-app /research page. Topic snapshots can be triggered too, gated by the brand's monthly snapshot credits (Trial 1, Growth 5, Scale 5, per active brand). Full prompt research runs are not exposed — they run daily on a schedule.

Tool	Description
get_latest_research	Most recent completed run with the full analytics payload. Canonical "what does my latest research show" lookup.
get_research_runs	List runs newest-first with visibility, mention rate, top competitors, and snapshot topic where applicable.
get_research_run	Full payload for one run: visibility, position distribution, competitor share, intent and focus breakdowns, best/missed prompts, per-prompt results.
get_research_credits	Topic snapshot credit usage for the current month.
run_research_snapshot	Trigger a topic snapshot (50 focused prompts). Async, 3-5 min; consumes one monthly credit (5/mo per active brand).

Crawler

Tool	Description
get_crawler_overview	Overview tab: hero stats, chart, setup state, top pages, and recent preview.
get_crawler_live	Live tab: activity, pages, or sessions with dashboard filters.
get_crawler_pages	Pages tab: pages, paths, or normalized bots.
get_crawler_detail	Drawer data for one page, path, or bot (kind + entity_id): verdict, pipeline, health, traffic, diagnostics.
get_crawler_access	Access tab: findings, bot matrix, robots.txt, llms.txt, submit-to-search.
preview_crawler_access_fix	Preview an Access fix without applying it.
send_crawler_verification_ping	Send the verification ping shown in the dashboard.
submit_crawler_to_search	Submit crawler pages to AI search via IndexNow.
get_crawler_submit_status	Read submit-to-search status and summary.

Audit & Recommendations

Read-only access to Trakkr's unified recommendation queue and site-audit data. These tools never trigger analysis or generation — they surface data the dashboard has already produced. Ideal for running an AI-search technical audit conversationally.

Tool	Description
get_actions	List recommended actions from the unified queue (audit, crawler, prompts, citations, etc.). Filter by source, category, action_type, lens, or quick_win.
get_action_stats	Aggregate counts by status, category, action_type, and source.
list_audits	List site audits for a brand (overall_score, issue counts, detected CMS).
get_audit_findings	Audit issues + flagged pages for the latest (or specified) audit. Includes has_llms_txt / ai_crawler_blocked / has_sitemap rollups.
list_page_analyses	List recent cached deep page analyses for a brand.
get_page_analysis	Cached deep analysis for one URL — AI diagnosis, citation verdict, failed checks, ready-to-paste schema, entities, bot visibility.

Actions

Tool	Description
get_results	List measured outcomes from completed work, including earned, no movement, drops, and results that could not be measured.
get_proof	Legacy compatibility alias for get_results.
manage_action	Act on one item in the recommendation queue: complete, dismiss, reopen, start, snooze, unsnooze, pin, unpin, assign, or note. One explicit verb per call, no destructive default. Editor access to the brand required.
run_diagnosis	Diagnose a search query across AI models in real-time.
get_diagnosis_result	Get diagnosis results, history, or usage quota.
generate_report	Generate an AI visibility report (executive, weekly, or full).
get_reports	List or retrieve generated reports.
export_data	Export data as JSON or CSV.
get_search_performance	Read the brand's connected Google Search Console data: property or page clicks, impressions, CTR, average position, top queries, and optional previous-period comparison.

Reddit

Tool	Description
get_reddit	Reddit monitoring data. Views: overview, feed, opportunities, thread, subreddits, triggers, analytics.
manage_reddit_subreddit	Add or remove monitored subreddits. Add uses the public API subreddit_name body field.
manage_reddit_trigger	Add or remove keyword triggers that drive Reddit scanning.
manage_reddit_opportunity	Dismiss an opportunity or mark it responded.
scan_reddit	Queue an on-demand Reddit scan.

Workflows

Tool	Description
get_workflows	Read workflow definitions, runs, run details, or templates.
manage_workflow	Pause, resume, soft-delete, or clone from a template. Delete maps to PATCH /workflows/{id} with status=deleted.

Activity

Tool	Description
get_changes	What moved for a brand since a timestamp: visibility, citations, competitors, rank, crawler signals, opportunities, reports. Returns a latest_seen cursor to pass back as since next time. A daily watch digest, not a live feed.
get_notifications	Read notifications by event_type, unread state, and lookback window.
mark_notifications_read	Mark specific notifications read or use mark_all=true for the brand.
manage_webhook	List, inspect, create, delete, or test webhooks. Create uses signing_secret and underscore event names such as visibility_changed.

Content

Tool	Description
get_knowledge	Read knowledge sources or aggregate stats. Views: sources, stats.
get_articles	Read articles. Views: list, detail. Status filters: draft, review, ready, published.
get_writing_style	Read writing style profile or source samples.
add_knowledge	Add a knowledge source from raw text or a URL. Processed async; poll get_knowledge until ready.
manage_knowledge	Delete or reprocess a knowledge source.
generate_article	Generate a brand article from a prompt, grounded in the brand's knowledge and voice. Async, consumes one article credit; poll get_articles until draft. Never publishes.

Agency

Tool	Description
list_brand_groups	List accessible agency brand groups with member brands.
compare_brands	Compare 2-10 brands across visibility, citations, and actions.
get_portfolio_actions	Highest-impact open actions across accessible brands, optionally filtered by group_id or quick_win.

Resources

Resources let a client pull Trakkr context into a chat with no tool call (for example @trakkr://brand/<id>/briefing in clients that support resources). Each one is a small, read-only, bounded view, so reading it never triggers analysis or compute.

Your brand (paste-ready briefings)

Markdown you can @-attach to any chat (writing a page, briefing an agency, steering another AI). Each reads like a brief, not a payload, and is summary-first and bounded.

Resource	Description
trakkr://brand	The index of your brands and their IDs, each linked to the briefings below. Start here to find a brand ID.
trakkr://brand/{brand_id}/brand-book	The star: paste-ready context so any AI describes the brand right. What it is, how AI frames it today, what to set straight, and the proof.
trakkr://brand/{brand_id}/snapshot	Latest AI visibility: the headline scores, a by-model table, and the prompts where you win and where you lose.
trakkr://brand/{brand_id}/citation-gaps	Prompts where AI cites a rival but not you, ranked, with who gets cited and what to publish to win them.
trakkr://brand/{brand_id}/prompts	The AI-search questions Trakkr is tracking for this brand, active and paused.

Your brand (live state, JSON)

Machine-shaped views for programmatic reads.

Resource	Description
trakkr://brands	The brands you can access, with the IDs that fill the per-brand templates below.
trakkr://brand/{brand_id}/briefing	Headline state: current visibility and trend plus the open-action snapshot.
trakkr://brand/{brand_id}/actions	The brand's top open actions, most impactful first.
trakkr://brand/{brand_id}/changes	What moved in the last week. The same digest as get_changes.
trakkr://brand/{brand_id}/latest-report	The most recent generated report (headline metrics and metadata).

Data & Research

Every rival shows you your own numbers. These resources expose Trakkr's public research on what actually moves the needle in AI search: the evidence behind getting cited. Each one reads as a sharp analyst briefing (a lead number, a tight table, a short "what to do"), fetched live from Trakkr's public /data studies and cached, so the figures stay current and the package never ships stale numbers. The data is public/CC, so attribution is welcome.

Resource	Description
trakkr://data	Index of the research briefings below, with a one-line hook for each.
trakkr://data/what-gets-cited	Which page types earn AI citations (blogs win, product and help pages lose), and owned vs third-party.
trakkr://data/crawler-personalities	What each AI bot actually fetches and reads, and how aggressively it crawls.
trakkr://data/llms-txt-truth	The honest null result: llms.txt shows no detectable citation lift.
trakkr://data/schema-advantage	How strongly structured data correlates with getting cited by AI.
trakkr://data/citation-decay	How fast AI citations fade: most last a single day.
trakkr://data/model-divergence	Where AI models disagree on who to recommend (they agree under half the time).
trakkr://data/playbook	The synthesis: the rules for getting cited in AI search, with links to the evidence.

Workflows (Prompts)

Prompts are server-provided slash commands. The moment Trakkr is connected they appear in your client's prompt menu (/weekly-review, /citation-gap-plan, and so on). Each one is a parameterised playbook: it chains the right tools and the trakkr://data research in the right order and shapes the result into a finished, scannable briefing, most important first. Arguments are optional. Leave a brand out and the workflow resolves it for you.

Prompt	What you get
weekly-review	Your Monday AI-search brief. Visibility and trend, who is gaining, the biggest citation gaps, and the easy wins, as five tight bullets. Args: brand, period (default 7d).
competitor-teardown	Where a rival beats you, where you beat them, and three moves to close the gap. Args: brand, competitor.
citation-gap-plan	The prompts you are losing, cross-referenced with the research on what actually gets cited, ending in a ranked plan of what to publish. Args: brand.
content-brief	A citation-optimised outline plus the on-page schema to include for one piece, ready to hand to a writing tool. Args: brand, topic.
setup-tracking	Onboarding in one conversation: confirm the brand, set its market, then propose a starter set of prompts to track and create them on your say-so. Args: domain.
trakkr-watch	A watch playbook. Pulls the latest changes for a brand, summarizes what moved in plain language, and offers next steps. On repeat checks, pass the last latest_seen cursor so you only see what is new. Args: brand_id, since.

Example Conversations

Once connected, you can ask your AI assistant things like:

• "How is my brand doing in AI search this month?"
• "Which competitors are gaining ground in ChatGPT?"
• "What content opportunities am I missing?"
• "Show me which pages get cited most by Perplexity"
• "Run a diagnosis on 'best project management tools'"
• "What does my visibility trend look like over the last 90 days?"
• "Show crawler overview for Nike"
• "What changed in Live this week?"
• "Preview crawler access fixes"
• "Send a verification ping"
• "Run a full AI search technical audit of my brand"
• "What are my top open actions right now?"
• "Show me every critical/high issue from the latest audit, and the worst-scoring pages"
• "What's the deep page analysis say about /pricing?"
• "Summarize my latest research run — visibility, top competitors, and missed prompts"
• "Run a topic snapshot on 'enterprise pricing pages' and tell me when it's done"
• "What Reddit threads should I jump into this week?"
• "List my active workflows and what fired in the last seven days"
• "/weekly-review Nike" for a five-bullet Monday brief, or "/citation-gap-plan" for a ranked plan of what to publish
• "/competitor-teardown Nike Adidas" to see exactly where a rival is winning and the three moves to close it
• "Keep an eye on Nike and tell me if anything moves" (uses the trakkr-watch prompt)
• "What changed for my brand since yesterday?"
• "Show unread visibility_drop notifications and mark them all as read"
• "Wire a webhook to https://example.com/hook for visibility_changed and citation_gained"
• "Compare these brands on visibility, citations, and actions"
• "Using Trakkr's research, what page types actually get cited in AI search?" (reads trakkr://data/what-gets-cited)
• "Does llms.txt help me get cited?" (reads trakkr://data/llms-txt-truth)
• "Give me the AI-search playbook for getting cited" (reads trakkr://data/playbook)

Requirements

• Python 3.10+
• A paid Trakkr account with an MCP connect token from Settings > Developer

Documentation

Full API documentation: trakkr.ai/learn/api

MCP setup guide: trakkr.ai/learn/api/mcp

License

MIT