Let your AI agents do compliance.

Connect the AI assistant your team already uses (Claude, Copilot, ChatGPT) to your firm's compliance records, with every action governed and logged.

It searches your archive, files reviews, and runs workflows on your behalf, all under your own permissions. Half the lookups a CCO does in a day get a grounded, cited answer, with the same audit trail and evidence as the web app.

Book a demo →

Claude Desktop · RegFin MCP

# Avery asks her assistant, in plain English You: Summarize the Henderson family's open trade exceptions and the comms that touched them this quarter. # the assistant calls RegFin's MCP tools for her → regfin.household.show(id="hh_henderson") ✓ → regfin.exceptions.list(status="open") ✓ 3 open → regfin.comms.search(since="quarter") ✓ 18 comms Assistant: 3 open exceptions — 2 high (TSLA concentration, net-margin leverage), 1 medium (cash drag). 18 related comms; an Apr-9 advisor email pre-discloses the TSLA add. Every record cited. every tool call logged · via=mcp · agent=claude

Bring your own AI

Hook it up to
whichever AI you use.

RegFin runs a standard Model Context Protocol server. Any MCP-compatible assistant connects in seconds. No lock-in to one vendor's model, no custom integration to maintain.

Speaks MCP with Claude Microsoft Copilot ChatGPT any MCP client

~/.config/claude/mcp.json Connected

// add RegFin · ~10 seconds, one block

{

"mcpServers": {

"regfin": {

"url": "https://mcp.regfin.com/sse",

"auth": "oauth",

"scopes": ["records.read", "filings.write"]

}

session ready · 50 tools · 4 resource types

→ regfin.search("Form PF deadline")

via=mcp · agent=claude · as=cco

← 3 records · 17 CFR §279.9

→ regfin.draft_review(obl_4821)

via=mcp · agent=claude · as=cco

← draft saved · evidence pack #4 sealed

⌘ MCP Server

Claude, Copilot, ChatGPT, and any MCP client

Add RegFin as a Model Context Protocol server. Your assistant can search the archive, read records, draft and submit reviews, and trigger workflows, all bound to your user identity, all logged.

→Any model: Claude, Microsoft Copilot, ChatGPT, or any assistant that speaks MCP. Switch models any time.
→50+ tools: search, read, write, attest, and trigger, exposed as MCP tools.
→Resources: records, comms, vendors, and exceptions exposed as MCP resources.
→Identity-bound: every agent call acts as you and inherits your permissions; it can't do anything you can't.

◆ Why it matters

A compliance copilot, on your terms

Half the questions a CCO answers in a day are simple lookups with a regulatory answer. Let the AI your team already uses handle those, grounded in your records, not the open internet.

→No new app to learn: it lives inside the assistant your team already opens.
→Grounded answers: every response is built from your archive, policies, and records, with citations.
→Audit-tagged: every action is logged with via=mcp · agent=claude so reviews stay auditable.
→No vendor lock-in: swap or add models without touching your compliance setup.

A working set, not a view-only client.

regfin.search(query)

Full-text across everything

Comms, documents, vendor records, trades. The same engine as web search, returned to your assistant with citations.

regfin.record.show(id)

Open any record

Pull the structured record for a comm, a doc, a household, or a vendor so the agent can reason over it.

regfin.review.submit

File a marketing review

Submit a draft for AI screen + CCO approval. Returns the review id; the agent can poll or wait for the outcome.

regfin.delivery.start

Kick off a campaign

Run an ADV / IMA / privacy delivery to an audience query. Returns the campaign id with streamable status.

regfin.exceptions.list

Triage trade exceptions

Filter by household, rule, severity, or date. Acknowledge, escalate, or override with a reason, under your identity.

regfin.vendor.attest(id)

Walk a vendor DDQ

Step through a vendor due-diligence questionnaire. Counts as a binding attestation under your identity.

regfin.form.export(year)

Pull a Form ADV diff

Return the current vs. last-filed ADV for the agent to summarize or reconcile.

regfin.events.tail(type)

Stream events

Subscribe to the queue (new exceptions, approvals, comms) so an always-on agent can react in real time.

regfin.audit.log(actor)

Pull the audit trail

The full log of what was done, by whom, when, and via what surface, returned for the agent to analyze.

Real recipes

What people actually
ask their AI to do.

The same assistant your team already opens. Behind each prompt, RegFin's MCP tools do the work across your compliance program: bound to your identity, grounded in your firm's records, and logged for your next SEC exam.

The MCP server acts on your behalf and inherits your permissions. Every action flows through the same permission checks as the web app and lands in the same audit log.

Inherits your permissions

The MCP server cannot exceed your web permissions. An advisor's agent can't approve marketing pieces; a marketer's agent can't see another's queue.

Step-up for sensitive actions

Approvals, signatures, and policy edits trigger a step-up confirmation on the linked mobile app. Even an agent can't bypass it.

Audit-tagged separately

Every action carries a via tag (web · mobile · mcp · api · agent_name) so reviews can isolate "what did the agent do" from "what did the human do."

Tokens, scoped & revocable

For non-interactive use, generate scoped tokens (e.g. read-only-archive, vendor-attest-only). Tokens carry their own audit identity and can be revoked instantly.

Guardrails on every agent

Limits on how much, and how fast, any agent can act, plus automatic checks for unusual behavior. A runaway agent is paused and you're alerted before it can do harm.

Your conversations stay private

RegFin never stores your prompts or chats. It keeps only the resulting actions, with provenance.

Common questions.

Compliance ops, first. The MCP server turns the AI assistant your team already uses into a compliance copilot. An advisor or analyst running Claude or Copilot with RegFin connected gets grounded, cited answers for half their daily lookups, with no new app to learn. The API page covers the developer-platform-integration case.

REST API is for system integration: your portfolio system calls us to push a trade, our system calls Salesforce to write an exception. MCP is for AI integration: Claude (or another agent) calls us with the user's identity to assist that user. Different audiences, different auth flows. Both exist; both are first-class.

It can call the wrong tool the same way a person can click the wrong button, so we treat it the same way. Every action lands in the audit log with via=mcp · agent=claude. Step-up authorization is required for irreversible actions. And the agent acts on the user's behalf and inherits their permissions. It can't do anything the person couldn't do in the web app.

Yes. Every action carries the surface tag. Examiners get to see "approval submitted via MCP by Avery Chen on 2026-04-12 at 14:32:01 UTC, with step-up confirmation from device id …". This is by design: more provenance is better, not worse, in audit.

Yes. The MCP tool catalog and the OpenAPI spec for the underlying calls are both open and published, so your security team can run any verification they like before connecting an agent. Internal pen-test reports available under NDA on request.

Yes. The underlying API is documented and stable. Some firms write a thin custom MCP layer that combines RegFin tools with their own data sources (custodian feed, OMS, internal CRM). Reach out and we'll send the integration guide.

Give your AI the compliance record.

See the assistant your team already uses connected to a sample firm's records: grounded, governed, and audited from the first question.