Proofsource and OpenAI agents: your agents, our evidence

Build agents on OpenAI that know what AI says about your brand. Pass Proofsource as a remote MCP tool with a workspace API key and your agent can read visibility, competitors, citations and fix opportunities, auto-approving only the read tools while anything that writes still waits for a yes.

Connect OpenAI agents See it in Proofsource

Live

Visit platform.openai.com ↗

app.proofsource.co/settings/mcp

Connect AI apps, OpenAI

{
  "type": "mcp",
  "server_label": "proofsource",
  "server_url": "https://mcp.proofsource.co/mcp",
  "headers": { "Authorization": "Bearer ps_live_..." },
  "require_approval": { "never": { "tool_names": ["get_visibility", "get_competitors"] } }
}

Replace the key with your own. Each key can make 60 tool calls a minute and 30 writes an hour.

Sample data

The OpenAI card with the tool block and which tools auto-approve.

Proofsource and OpenAI agents, in short

The facts, ready to quote.

Stage: Live
Connects with: A remote MCP tool in the Responses API with a workspace API key as a Bearer header
Approvals: Read tools auto-approve; write tools ask
Limits: 60 tool calls a minute and 30 writes an hour per key

Why we love OpenAI agents

In the makers' own words, from their own pages. Proofsource builds on it, never around it.

Remote MCP in one tool definition

The API accepts remote MCP servers as a tool, so a model can reach outside systems with a short, declarative config.

From platform.openai.com

Approvals you control

Tool calls ask for approval by default, and you can relax that for chosen tools, which makes safe automation a setting rather than a rewrite.

From developers.openai.com

Agents you can build and ship

The platform offers an agents API, an agents SDK and the Responses API, with function calling underneath, to fit the shape of whatever you are building.

From developers.openai.com

How Proofsource uses OpenAI agents

Call Proofsource tools from your own OpenAI agents, auto-approving only the read-only ones.

Connect
One tool definition
Add an MCP tool with Proofsource's server URL, a label and a Bearer header holding your workspace key.
Know
Read tools run freely
Your agent can call list_brands, list_topics, get_visibility, get_competitors, get_citations, get_changes, list_fix_opportunities, and get_fix_verification without a prompt for approval.
Act
Writes wait
Creating a buyer question or marking a fix shipped requires approval, and a read-only key cannot write at all.
Prove
Close the loop in code
An agent can mark a fix shipped and later read its verification, so reporting runs without a person pulling numbers.

Set up OpenAI agents in 4 steps

Create a workspace API key
Go to Settings, API Keys. Make a read-only key for reporting agents and a write key only where an agent should track questions or mark fixes shipped.
Copy the tool snippet
Settings, Connect AI apps, For developers shows the OpenAI card with your server URL and the require_approval setting filled in.
Pass it in a Responses call
Add the snippet to the tools list, with the key in the Authorization header, and ask your agent a question about your visibility.
Rotate or revoke when needed
Delete the key to cut access at once. Each key is limited to 60 tool calls a minute and 30 writes an hour.

app.proofsource.co/settings/api-keys

API Keys

Name	Access	Last used
Weekly report agent	Read	Today
Fix tracker agent	Read and write	Yesterday
Board summary agent	Read	Nov 1

Create key

Sample data

A read-only key for reporting agents, and a write key for one that ships fixes.

What you unlock with OpenAI agents

Agents that know your brand

Your own OpenAI agents read visibility, competitors and citations.

Read-only by default

Only read tools auto-approve; writes always wait.

Reports on autopilot

Schedule an agent to write your weekly AI visibility summary.

Verification in the loop

Mark a fix shipped and read back its before and after.

Fair limits

60 tool calls a minute and 30 writes an hour per key.

Per-key control

Each agent gets its own key you can revoke on its own.

Google AI Overviews

Google AI Mode

Questions about Proofsource and OpenAI agents

How does my OpenAI agent call Proofsource?

Add a remote MCP tool to a Responses API call, with Proofsource's server URL and a workspace API key as a Bearer header. The For developers panel in Settings gives you the snippet ready to paste.

Which tools run without approval?

Only the read-only ones. The snippet auto-approves the read tools, and create_prompt and mark_fix_shipped still require approval. A read-only key cannot call them at all.

What access does the key give?

It reaches one Proofsource workspace and the brands in it, at the level you chose when you made it. Create separate keys for separate agents so each can be revoked alone.

How do I disconnect an agent?

Delete its API key under Settings, API Keys. Its next tool call is refused.

What does Live mean for this integration?

It is generally available on the Connect AI apps screen in Settings.

How does this help my AI visibility?

Agents can watch your visibility, flag gaps and track fixes without a person pulling reports, so changes in ChatGPT, Perplexity and the other engines get noticed early.

Which AI engines does Proofsource check?

Up to 10 depending on your plan: ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot, Grok, DeepSeek and Meta AI. The free scan and trial run four.