Developers

Developer tools · MCP

Use Tritonium from your AI agent

Tritonium runs a remote Model Context Protocol (MCP) server. Sign your AI agent in with your Tritonium account, or give it a scoped API key, and it can read your apps, reviews, insights and rating forecasts, and prepare reply drafts that wait for your team's approval.

Endpoint

Connection details

Every client connects to the same URL. Clients that support MCP sign-in need nothing else; the others send a Tritonium API key.

Endpoint
https://api.tritonium.com/mcp
Transport
Streamable HTTP with JSON responses. There is no SSE stream and no session.
Protocol versions
MCP 2026-07-28, and the earlier initialize-based revisions 2025-03-26 through 2025-11-25, on the same URL.
Sign-in
OAuth 2.1 with PKCE, as the MCP authorization specification describes. A request without credentials gets a 401 response that points to https://api.tritonium.com/.well-known/oauth-protected-resource/mcp. The authorization server's metadata is at https://api.tritonium.com/.well-known/oauth-authorization-server.
API key
A Tritonium API key, sent as Authorization: Bearer <key> or in an X-API-Key header. Dashboard session tokens are not accepted.
Health check
https://api.tritonium.com/mcp/health

Quick start

Set up in three steps

Add Tritonium to your agent, sign in and approve what it may do, then ask your agent a question.

  1. Add Tritonium to your agent

    Each setup below adds the server by its URL. The first time the agent connects, it asks you to sign in to Tritonium. Every client also shows how to use an API key instead.

    Claude Code

    Terminal
    claude mcp add --transport http --scope user tritonium https://api.tritonium.com/mcp

    Then run /mcp in Claude Code, choose tritonium and sign in. From a terminal, claude mcp login tritonium does the same. Over SSH, add --no-browser: open the link it prints on your own computer, and after you approve, paste the address your browser ends up on back into the terminal.

    Use an API key instead

    First create a key and store it in the TRITONIUM_API_KEY environment variable, as described under API keys.

    Terminal
    claude mcp add --transport http --scope user tritonium https://api.tritonium.com/mcp --header 'Authorization: Bearer ${TRITONIUM_API_KEY}'

    The single quotes keep ${TRITONIUM_API_KEY} literal in the saved configuration; Claude Code reads the variable when it connects. Check the connection with claude mcp list.

    Codex

    Terminal
    codex mcp add tritonium --url https://api.tritonium.com/mcp

    Codex notices that the server supports sign-in and opens your browser. It asks for every permission the server offers, so untick the ones your agent doesn't need before you approve. To sign in again later, run codex mcp login tritonium.

    Use an API key instead

    First create a key and store it in the TRITONIUM_API_KEY environment variable, as described under API keys.

    Terminal
    codex mcp add tritonium --url https://api.tritonium.com/mcp --bearer-token-env-var TRITONIUM_API_KEY

    Codex reads the key from the TRITONIUM_API_KEY environment variable.

    Cursor

    Add this to .cursor/mcp.json in a project, or to ~/.cursor/mcp.json for all your projects. Then approve the server and sign in with the Cursor CLI:

    mcp.json
    {
      "mcpServers": {
        "tritonium": {
          "url": "https://api.tritonium.com/mcp"
        }
      }
    }
    Terminal
    cursor-agent mcp enable tritonium
    cursor-agent mcp login tritonium

    The Cursor CLI waits for your browser on http://localhost:8787/callback, so port 8787 must be free on the computer where it runs. Over SSH, forward that port with ssh -L 8787:localhost:8787 <host>; the Cursor CLI prints the same hint.

    Use an API key instead

    First create a key and store it in the TRITONIUM_API_KEY environment variable, as described under API keys.

    Add this to .cursor/mcp.json in a project, or to ~/.cursor/mcp.json for all your projects.

    mcp.json
    {
      "mcpServers": {
        "tritonium": {
          "url": "https://api.tritonium.com/mcp",
          "headers": { "Authorization": "Bearer ${env:TRITONIUM_API_KEY}" }
        }
      }
    }

    Cursor fills in ${env:TRITONIUM_API_KEY} from your environment. If Cursor reports an authentication error or offers to sign in, it could not use the key: check that the variable is set and the key is still active.

    GitHub Copilot CLI

    Terminal
    copilot mcp add --transport http tritonium https://api.tritonium.com/mcp

    Then start copilot and run /mcp auth tritonium. Sign-in starts only in an interactive session, not with copilot -p. If Copilot still says the server requires authentication afterwards, restart copilot: it then uses the stored sign-in, also with copilot -p.

    Use an API key instead

    First create a key and store it in the TRITONIUM_API_KEY environment variable, as described under API keys.

    Add this to ~/.copilot/mcp-config.json:

    mcp-config.json
    {
      "mcpServers": {
        "tritonium": {
          "type": "http",
          "url": "https://api.tritonium.com/mcp",
          "headers": { "Authorization": "Bearer ${TRITONIUM_API_KEY}" },
          "tools": ["*"]
        }
      }
    }

    Copilot fills in ${TRITONIUM_API_KEY} from your environment.

    Grok CLI

    Terminal
    grok mcp add tritonium https://api.tritonium.com/mcp

    Then start grok, open /mcps, select tritonium and press i to sign in.

    Use an API key instead

    First create a key and store it in the TRITONIUM_API_KEY environment variable, as described under API keys.

    Add this to ~/.grok/config.toml:

    config.toml
    [mcp_servers.tritonium]
    url = "https://api.tritonium.com/mcp"
    bearer_token_env_var = "TRITONIUM_API_KEY"

    Grok reads the key from the TRITONIUM_API_KEY environment variable.

    Antigravity

    Terminal
    agy mcp add tritonium https://api.tritonium.com/mcp

    Then start agy, open /mcp, choose tritonium and sign in. After you approve, your browser shows an Antigravity page with a code: paste that code into agy. The stored sign-in also works with agy --print.

    Use an API key instead

    First create a key and store it in the TRITONIUM_API_KEY environment variable, as described under API keys.

    Add this to ~/.gemini/config/mcp_config.json and replace the placeholder with your key.

    mcp_config.json
    {
      "mcpServers": {
        "tritonium": {
          "serverUrl": "https://api.tritonium.com/mcp",
          "headers": { "Authorization": "Bearer trtn_live_your_key" }
        }
      }
    }

    Antigravity does not expand environment variables in headers, so the key itself is stored in this file. Keep the file private (chmod 600) and out of version control.

    OpenClaw

    Terminal
    openclaw mcp set tritonium '{"url":"https://api.tritonium.com/mcp","transport":"streamable-http","auth":"oauth"}'
    openclaw mcp login tritonium

    OpenClaw waits for your browser on http://127.0.0.1:8989/oauth/callback. If the browser cannot reach it, for example over SSH, finish with the openclaw mcp login tritonium --code <code> command that OpenClaw prints.

    Use an API key instead

    First create a key and store it in the TRITONIUM_API_KEY environment variable, as described under API keys.

    Terminal
    openclaw mcp set tritonium '{"url":"https://api.tritonium.com/mcp","transport":"streamable-http","headers":{"Authorization":"Bearer ${TRITONIUM_API_KEY}"}}'

    The single quotes keep ${TRITONIUM_API_KEY} literal in the saved configuration; OpenClaw fills it in from your environment when it loads the server.

    Other MCP clients

    Clients that support MCP authorization need only the URL https://api.tritonium.com/mcp. They find the sign-in on their own from the server's 401 response. The official MCP Python SDK's built-in OAuth support works without any Tritonium-specific settings.

    Clients that cannot sign in can use an API key if they support remote Streamable HTTP servers with custom headers: send the header Authorization: Bearer <key>.

  2. Sign in and approve

    The first time your agent connects, it opens Tritonium in your browser. Sign in, check which app is asking and where it sends access, choose what the agent may see and do, then approve. The agent then works for you, within the permissions you chose.

    Every workspace member can approve

    • Your apps apps:read See your apps and their details.
    • Reviews and replies reviews:read Read reviews, replies and reply drafts.
    • Insights and analytics analytics:read Read insights, ratings, forecasts and activity.
    • Triggered alerts alerts:read Read the alerts your workspace has received.

    Admins and owners can also approve

    • Reply drafts reviews:write Write reply drafts for your team to review. The agent can't publish them.
    • Start app analysis and forecasts apps:write Start AI analysis or generate rating or sentiment forecasts, or cancel an analysis still waiting to run.
    • Webhook settings webhooks:manage Read webhook settings and delivery logs, without changing them.
  3. Ask your agent

    Try a prompt like one of these. Your agent chooses the tools, and any reply it drafts waits for approval in Tritonium.

    • Which of my apps got the most one-star reviews this week, and what are people complaining about?
    • Summarize the latest AI analysis for my iOS app and list the top feature requests.
    • Our rating dropped last week. Compare the rating history with recent reviews and tell me what changed.
    • Draft replies to the five most recent negative reviews so I can approve them in Tritonium.

API keys

Use an API key instead

API keys still work with every client on this page. Use one when your client can't sign in, for scripts and CI, or with the Tritonium CLI.

Create an API key

A workspace admin or owner creates keys in the Tritonium dashboard under Settings → API keys. Give the key only the scopes your agent needs, then make it available as an environment variable where your client runs:

Terminal
export TRITONIUM_API_KEY="trtn_live_…"

Then open Use an API key instead in your client's setup above.

For scripts and CI, the Tritonium CLI calls the same tools and prints JSON. See the CLI guide.

Reference

Tools

The server has 21 tools. Your agent sees only the tools its permissions allow, and a :write scope also grants read access in the same domain.

Tritonium MCP tools with the scope each one needs
Tool What it does Scope
get_workspace_context Your workspace, the granted scopes, and which tools the agent can use. Always available
list_apps Apps with review counts, sync and analysis status, and whether new reviews are still coming in. apps:read
get_app One app's details, live sync and analysis status, and whether new reviews are still coming in. apps:read
search_reviews Filter one app's reviews by text, rating, sentiment (from the star rating unless labelled), language and date. reviews:read
get_review One review in full, with its translation and any developer response. reviews:read
list_reply_drafts AI reply drafts waiting for approval. reviews:read
get_reply_draft One reply draft by ID, with its current status, text and review identifiers. reviews:read
generate_reply_draft Writes an AI reply draft for human approval. It is never published. Writes reviews:write
start_app_analysis Queues a new AI analysis of an app's reviews. Supports a dry run. Writes apps:read (dry_run) / apps:write (start)
get_analysis_run One AI analysis run: its status, stage, progress and any failure. apps:read
list_analysis_runs An app's AI analysis runs, newest first, including runs started after review syncs. apps:read
cancel_analysis_run Cancels an AI analysis run that is still waiting to run. A run that is already being processed continues. Writes apps:write
get_app_insights The latest AI analysis: summary, key issues and strengths, feature requests, bugs and topics. analytics:read
get_rating_history Daily or weekly star-rating trend. analytics:read
get_cached_rating_forecast Read a cached AI rating forecast without starting generation. analytics:read
get_rating_forecast AI rating forecast for the next 7 to 90 days. Writes The first request may start generating the forecast. analytics:read + apps:write
list_feature_requests Feature requests extracted from reviews. analytics:read
list_alerts Alerts raised for the workspace. alerts:read
list_webhooks Webhook endpoints and delivery counts. Secrets are never returned. webhooks:manage
list_webhook_deliveries Recent delivery attempts for one webhook. webhooks:manage
list_activity The workspace activity feed. analytics:read

Also on the server

  • Resource tritonium://guides/agents: a guide for agents working with Tritonium.
  • Prompts triage_negative_reviews and investigate_rating_change: starting points for triaging negative reviews and investigating a rating change.

Permissions

You stay in control

An agent works inside the permissions you gave it, and the decisions that reach your customers stay with your team.

Scoped, revocable access

A signed-in agent or an API key always acts as a workspace member limited to its permissions. Admins and owners disconnect agents under Settings → API keys → Connected agents and revoke or rotate keys under Settings → API keys. Members disconnect their own agents under Settings → Security → Connected agents.

Nothing is published automatically

Review and approve reply drafts in the dashboard. Store publishing requires a supported connection and store credentials; you can also copy the reply into the store console to publish manually.

Visible usage

The API keys tab shows when each key was last used and how many requests it has made. Connected agents shows who connected each agent and when it was last used.

What agents cannot do

These actions stay with signed-in admins and owners in the Tritonium dashboard. No API key and no permission you approve lets an agent take them:

  • Publish, approve or reject replies in the app stores
  • Connect, delete or re-sync apps
  • Acknowledge alerts or edit alert rules
  • Create or test webhooks
  • Exports, billing, team and credential settings

Limits

Good to know

  • The Tritonium CLI accepts API keys or supplied OAuth access tokens; it does not sign in or refresh tokens.
  • Dashboard session tokens are not accepted. Sign the agent in or use an API key.
  • Each API key and each signed-in agent has its own budget, shared with the REST API: 600 requests a minute, and 60 new AI drafts, analyses or forecasts every 10 minutes. Requests are also limited per IP address. When a tool returns RATE_LIMITED, check retry_after_seconds before trying again.
  • AI analysis and reply drafts use your plan's AI capacity. Estimates use written reviews, not all store ratings. Accuracy against future displayed store ratings has not been established. Model ranges are not calibrated confidence intervals.
  • Browser-based clients can connect only from Tritonium's own origins.
  • Grok Bot, xAI's hosted agent, has not been tested with Tritonium.

Structured errors

Failures come back as tool results with a stable error.code, a retryable flag and retry_after_seconds, so an agent knows whether and when to try again.

Error codes

  • INVALID_ARGUMENT
  • AUTHENTICATION_FAILED
  • PERMISSION_DENIED
  • SUBSCRIPTION_INACTIVE
  • NOT_FOUND
  • CONFLICT
  • RATE_LIMITED
  • UPSTREAM_UNAVAILABLE
  • UPSTREAM_TIMEOUT
  • INTERNAL_ERROR

Tested with

The client setups on this page were tested on with Claude Code 2.1.277, Codex CLI 0.155.0, Cursor CLI 2026.09.15, GitHub Copilot CLI 1.0.86, Grok CLI 1.0.34, Antigravity CLI 1.2.6, OpenClaw 2026.9.4, and MCP Python SDK 2.2.0. Each of them signed in to the production server.

Bring your app reviews to your agent

Connect your App Store and Google Play apps in Tritonium, then add the server to your agent and sign in.