Skip to content
trovance

Agent access (MCP)

Use Trovance from your AI agent.

Trovance runs an MCP server. Give your coding agent or assistant a key, and it can read your answers, queue and drafts, and file briefs. With an operate key it can also act for you: decide cards, edit drafts and publish them.

Client setup checked against each vendor’s documentation on October 18, 2026

Before you connect

One endpoint, one key per brand

  • The endpoint

    https://app.trovance.ai/api/mcp/mcp, with mcp twice. https://app.trovance.ai/api/mcp on its own returns not found.

  • The transport

    Streamable HTTP. Trovance doesn’t offer SSE, so don’t choose it, and don’t add /sse to the address.

  • The key

    Sent in the Authorization header, as Bearer YOUR_TROVANCE_KEY. Trovance takes API keys only: no OAuth sign-in, and no key in the address.

  • One brand per key

    A key belongs to one brand, so an agency mints one for each client brand. Every plan includes agent access: Starter, Growth and Agency.

Building your own agent? The discovery document is at https://app.trovance.ai/.well-known/mcp.json.

Create a key

Mint a key in Settings

  1. Sign in as a workspace Owner or Admin. Editors and viewers don’t see agent access.

  2. Open Settings → Connections → Advanced, and find the Agent access card.

  3. Choose Operate or Suggest, then select Mint operate key or Mint suggest key. Choose Suggest unless the agent needs to publish or act: an operate key can change and publish product state.

  4. Copy the key straight away: it’s shown once. Keep it in your agent’s settings or an environment variable, never in a file you commit.

  • Rotate or revoke a key from the same card. Any agent using a revoked key loses access on its next call. Revoking can’t be undone: you mint a new key instead.
  • A key also stops working for good if the person who minted it stops being an Owner or Admin, leaves the workspace, or the trial runs out without a paid plan. Upgrading won’t bring it back; mint a new one.
  • On a new trial, before the card is added at checkout, a key can read but can’t file briefs or act.
  • On a Starter workspace that began on the card-at-checkout trial, a key minted by an Admin who isn’t the owner can read but can’t file briefs or act.

What a key can do

Read, suggest, operate

A key carries the role its minter chose, and every action an agent takes with it is attributed to that person.

Read

Every key

  • Who the key belongs to, and what it may do right now
  • The brand, and the Brand Core that steers drafting: positioning, description and audiences
  • How many answers named or cited your brand, by engine for the latest week and in total for each of the last eight weeks (these counts include Claude memory, so they won’t match the app’s share of answer)
  • Recent answers, the first 600 characters of each word for word, with whether your brand was named in it
  • The queue of recommendation cards, with their evidence
  • A search of the Brain: voice lessons, learnings, proof points, competitor notes, brand facts and imported notes
  • Pixel traffic: visits by traffic class, the AI engines sending them, top pages and goals
  • The topic map: pillars, topic counts and published coverage
  • Your content: titles, status, word counts, schedule dates and live addresses
  • One piece in full: its Markdown, content score and status, and whether someone is editing it right now

Suggest

Every key

  • File a brief as a proposed card in your queue. Nothing runs until someone accepts it, in the app or with an operate key.

Operate

Operate keys only

  • Accept or dismiss a proposed card; accepting sends it to drafting, as Accept does in the app
  • Replace a draft’s text and title, stepping back if someone is editing it
  • Set or clear a draft’s planned publish date
  • Publish a draft, and receive its deploy-ready files
  • Report where a published page went live, so Trovance can verify it
  • Add a proof point, competitor note or brand fact to the Brain
  • Preserve a source in the Brain, exactly as given
  • Add a claim to the claim bank, marked as needing verification until a person grades it
  • Comment on a draft, in the minter’s name, marked as via your agent
  • Run an analysis cycle now, filing fresh evidence-backed cards

An operate key acts with the authority of the person who minted it. It can change your queue, rewrite drafts and publish them in Trovance. Mint one only for an agent you’d trust with those controls, and a suggest key when the agent only needs to read and suggest.

Every key can be revoked. Under the key card, Agent activity lists every call an agent ran against your brand, with the state it answered with and how long it took. It never keeps what a call returned.

Set up your client

Connect your agent

Put your key where a snippet says YOUR_TROVANCE_KEY, or set it as TROVANCE_API_KEY where a client reads it from the environment. No Trovance tool is marked read-only, so a client that asks before an agent writes will ask before it reads too.

Claude Code

Supported

Set TROVANCE_API_KEY to your key in your shell first, so the key itself stays out of your shell history, then add the server from your project’s folder:

claude mcp add --transport http trovance https://app.trovance.ai/api/mcp/mcp \
  --header "Authorization: Bearer $TROVANCE_API_KEY"

That saves it in Claude Code’s own settings for this project, not in the project’s files. Add --scope user to use it in every project. To share it with your team, put it in the project’s .mcp.json, with the key in an environment variable rather than in the file:

.mcp.json
{
  "mcpServers": {
    "trovance": {
      "type": "http",
      "url": "https://app.trovance.ai/api/mcp/mcp",
      "headers": {
        "Authorization": "Bearer ${TROVANCE_API_KEY}"
      }
    }
  }
}

Set TROVANCE_API_KEY to your key in the shell that starts Claude Code, then check the connection with claude mcp list.

Cursor

Supported

Add this to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for every project:

.cursor/mcp.json
{
  "mcpServers": {
    "trovance": {
      "url": "https://app.trovance.ai/api/mcp/mcp",
      "headers": {
        "Authorization": "Bearer ${env:TROVANCE_API_KEY}"
      }
    }
  }
}

Set TROVANCE_API_KEY to your key in your shell profile or system environment, where the Cursor app can see it. Cursor asks before it uses a tool, and on Enterprise an admin’s MCP allowlist may need to include app.trovance.ai.

VS Code and GitHub Copilot

Supported

Add this to .vscode/mcp.json in your workspace. VS Code asks for the key the first time the server starts, and stores it securely:

.vscode/mcp.json
{
  "inputs": [
    {
      "type": "promptString",
      "id": "trovance-api-key",
      "description": "Trovance API key",
      "password": true
    }
  ],
  "servers": {
    "trovance": {
      "type": "http",
      "url": "https://app.trovance.ai/api/mcp/mcp",
      "headers": {
        "Authorization": "Bearer ${input:trovance-api-key}"
      }
    }
  }
}

On Copilot Business or Enterprise, your organization’s policy for MCP servers in Copilot has to allow it.

Codex

Supported

Add this to ~/.codex/config.toml, or to .codex/config.toml in a trusted project:

config.toml
[mcp_servers.trovance]
url = "https://app.trovance.ai/api/mcp/mcp"
bearer_token_env_var = "TROVANCE_API_KEY"

Set TROVANCE_API_KEY to your key in the environment that starts Codex, then check with codex mcp list. If the variable isn’t set, Codex can connect without a key, and Trovance turns it away.

Windsurf (now Devin Desktop)

Supported

Windsurf is now called Devin Desktop, and its documentation shows how to add a remote server with a static key in headers, though not with an Authorization header. Add Trovance as its MCP documentation describes, with the endpoint above and Bearer YOUR_TROVANCE_KEY as the value of an Authorization header. The key then sits in Windsurf’s own settings file: keep that file out of any repository, and use a suggest key unless the agent needs to publish.

Claude Desktop and claude.ai

Only with Request headers

Supported only where Anthropic’s Request headers beta is enabled: the custom connector needs it to send your key. It’s available to a limited set of organizations, and if the Add custom connector dialog has no Request headers section, yours doesn’t have it yet.

  1. In Customize, open Connectors and choose Add custom connector.

  2. Name it Trovance, and enter https://app.trovance.ai/api/mcp/mcp as the MCP server URL.

  3. Under Authentication, choose No sign-in.

  4. Under Request headers, pick authorization and enter Bearer YOUR_TROVANCE_KEY as its value, space included. Mark it Required, then select Add.

On Team and Enterprise, an Owner adds it under Organization settings → Connectors → Add → Custom → Web, and members connect it from Customize → Connectors. The key is shared there, so everyone who uses the connector acts with the minter’s authority: use a suggest key. Claude can’t show or edit the key afterwards; to rotate it, remove the connector and add it again.

ChatGPT

Not supported today

Not supported today. ChatGPT’s custom MCP connections offer OAuth, no authentication, or a choice of the two, with no field for an API key or bearer token. Trovance connects only with a key, so ChatGPT has no way to send it.

Walkthrough

Publish to a Next.js or Astro site with your coding agent

If your site is a Next.js or Astro project in Git, a coding agent such as Claude Code, Cursor or Codex can carry an approved piece from Trovance into a pull request. Trovance never touches your repository: the agent does, and your team reviews the change. It takes an operate key from someone who can publish.

  1. A person picks the piece.

    Decide which draft is ready to ship, then ask the agent for it. It lists your content and reads that draft in full, including whether someone is editing it right now.

  2. The agent publishes it in Trovance.

    With an operate key, the agent publishes the draft. Trovance marks it published and returns the full HTML, with Article structured data, FAQ structured data when the draft has an FAQ, and author markup, plus an llms.txt for your brand that lists your earlier published pieces with a live address. This piece isn’t in that list yet. The draft it read already gave it the Markdown. Nothing is on your site yet.

  3. It writes the article into your repo.

    The agent adds the piece where your site keeps its content, as Markdown or as HTML, whichever your site uses, without changing its words. If your site has an llms.txt, it adds a line for the new piece. Trovance’s HTML gives the piece a canonical address under /blog/ on your domain, so if your posts live at another path, the agent should point it at the real one.

  4. It opens a pull request.

    Your team reviews the change, and merges it or doesn’t. A person decides what ships. Settle the words in Trovance before the agent publishes: verification looks for the exact saved text, so a change made only in the pull request won’t verify, and a closed pull request leaves the piece marked published in Trovance.

  5. After the merge, it reports the live address.

    Once the page is live, someone tells the agent, and it reports the address to Trovance, on your own domain or a subdomain. Trovance fetches the page and looks for the saved text, in order: that’s what verified means.

A prompt to start from

Find the Trovance draft titled "..." and publish it in Trovance. Add the article to our blog's content folder as Markdown without changing its words, add a line for it to our llms.txt, and open a pull request. When I tell you the pull request is merged and the page is live, report its URL to Trovance.

Put Trovance in your agent.

Get understood. Get cited. Get chosen.

See how AI talks about your brand.

Text Trovance →

Opens iMessage with your note written.

You choose to send it. Trovance replies, and sends the updates you ask for, over iMessage through our messaging provider. Reply STOP anytime. Tapping or sending never bills you.

How Trovance measures →