For FCA-regulated firms
See what your AI agents do. Control what they can reach.
ContextGate is one MCP connection you add to the AI apps your staff already use. Every tool call those apps make through it is logged and checked against your policies. Risky calls are blocked, and personal data in tool results is redacted before the AI app sees it.
Two-week audit · fixed fee, quoted on the scoping call · ContextGate itself is free
The three questions
When an AI agent touches customer data, someone senior will be asked three questions.
Which agents can reach customer data?
The Agents page lists each AI app and API key using ContextGate, and who signed it in. Switch a tool off and it's hidden from those apps and refused if called.
What did that agent do, and should it have?
Every tool call that goes through ContextGate is logged with the request, the redacted response and whether it was allowed, warned or blocked. Governance checks block risky actions before they run.
What happens if we change AI vendor?
Policies, redaction and the activity log live in ContextGate, not in the AI app, so they work across AI apps that support MCP.
How it works
Connect.
Add each system your agents use, such as Salesforce, Slack, GitHub or any remote MCP server, as a connection. Then add ContextGate to each AI app as a remote MCP server and sign in. It takes minutes.
Govern.
Apply policies to every tool call: redaction of personal data including UK sort codes, account numbers, NI numbers and IBANs, blocking of calls that carry secrets, and plain-English governance checks that block or warn. Start from the UK templates.
Evidence.
Activity logs each tool call with the AI app, user, tool, arguments, redacted result and policy outcome. Attention lists the last seven days of blocked and warned calls in one place.
Frameworks you already answer to
ContextGate doesn't make you compliant. It helps you gather evidence for these frameworks.
SM&CR
Each AI app connection is tied to the person who signed it in, and every tool call is logged against the app and user that made it.
Consumer Duty
The UK FCA Consumer Duty policy template checks tool calls for vulnerable-customer signals, foreseeable harm and one-sided value claims, then blocks or warns.
UK GDPR
Personal data in tool results is replaced with placeholders before the AI app sees it. The log records which types of data were removed and how many, not the values.
AML, promotions and Open Banking
Templates for UK AML/KYC, UK Financial Promotions (FSMA s.21), UK Open Banking (PSD2/FCA PSRs) and UK Insurance Privacy. Each one is a starting point you can adjust.
The audit
AI Agent Risk Audit
- Duration
- 2 weeks
- Price
- Fixed fee, quoted on the scoping call (excluding VAT)
- Afterwards
- ContextGate itself is free to use. No card needed.
- What we do
- List the AI tools we can find in use across your firm; connect one to three of your MCP-capable AI apps and their tools to a ContextGate workspace; apply the UK policy templates and show you what gets blocked, warned on and redacted
- What you get
- A written report: the AI tools we found, what the connected apps reached, and what the UK templates blocked, warned on and redacted, with gaps mapped to SM&CR, Consumer Duty and UK GDPR, and a remediation plan
Audit timeline
Days 1–3 · Discover.
We list the AI tools we can find from exports you already have: identity-provider app lists, Microsoft 365 or Google Workspace OAuth grants, AI domains in proxy logs and AI subscriptions in expense data. Plus a short staff survey. Nothing to install.
Days 4–8 · Test.
We connect one to three of your MCP-capable AI apps, and the tools they use, to a ContextGate workspace and apply the UK policy templates. You see what gets blocked, warned on and redacted in Activity and Attention.
Days 9–10 · Report.
You get a written report mapped to SM&CR, Consumer Duty and UK GDPR, and a remediation plan.
Scope: ContextGate sees only the tool calls that go through it, not an AI app's built-in tools or chat. Discovery finds the AI tools in use; the test shows what the connected ones do.
A note from the founder
I built ContextGate after watching firms ship AI agents before anyone could say what those agents were allowed to do. If you're rolling agents out in a regulated firm, the audit is the fastest way to find out where you stand.
Frequently asked questions
Where is our data processed and stored?
ContextGate runs on Google Cloud in the US (us-central1). App integrations go through Composio, which holds the OAuth grants for the apps you connect. Governance checks run through OpenRouter, so the model running a check sees the full tool call, before any redaction. UK or EU hosting isn't available today. If it's a requirement for you, raise it on the scoping call and we'll tell you where it stands.
Which AI tools does it work with?
AI apps that support remote MCP servers. There are setup guides for Claude (web, desktop and Cowork), Claude Code, Cursor, VS Code, ChatGPT (developer mode), Codex and Gemini CLI. Any other MCP client, and your in-house agents, can connect with an API key.
Does it see everything our staff do with AI?
No. It sees the tool calls that go through it: which AI app called which tool, for which user, with what arguments, and what came back. It doesn't see an AI app's built-in tools or the chat itself. That's why the audit's discovery phase starts from your own exports.
Can it run in our own cloud?
Not today. ContextGate is a hosted service, and the audit uses it.
What does the audit need from us?
For discovery: exports you already have, such as your identity provider's app list, Microsoft 365 or Google Workspace OAuth grants, proxy logs and expense data, plus answers to a short staff survey. Nothing is installed. For the test phase: one to three AI apps that support MCP, the tools they use, and someone who can sign in to each.
What does it cost after the audit?
Nothing. ContextGate is free to use and needs no card. The audit is a separate, fixed-fee service.