Build with C4C: the regulation library, in your own tools.

Your AI knows the gist. Your tools need the regulator’s exact words, with the citation, the version in force and a way to prove the quote is real. Pick the path that fits how you build; each takes a few minutes.

1. Agents: connect the MCP server

https://mcp.claudeforcompliance.com/mcp works with any client that speaks MCP over HTTP. No key, no sign-up. Your agent gets the same tools a person gets in their chat: search, read a section verbatim, resolve a citation, and check a quote.

Claude Agent SDK (TypeScript)

import { query } from "@anthropic-ai/claude-agent-sdk";

for await (const m of query({
  prompt: "What does 24 CFR 203.402 allow for hazard insurance? Quote it.",
  options: {
    mcpServers: { c4c: { type: "http", url: "https://mcp.claudeforcompliance.com/mcp" } },
    allowedTools: ["mcp__c4c"],
  },
})) {
  if (m.type === "result") console.log(m.result);
}

OpenAI Agents SDK (Python)

import asyncio
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp

async def main():
    async with MCPServerStreamableHttp(
        name="c4c", params={"url": "https://mcp.claudeforcompliance.com/mcp"}, cache_tools_list=True
    ) as c4c:
        agent = Agent(
            name="Compliance research",
            instructions="Quote the regulator's exact words with the citation. Check each quote with verify_quote.",
            mcp_servers=[c4c],
        )
        result = await Runner.run(agent, "What does 24 CFR 203.402 allow for hazard insurance?")
        print(result.final_output)

asyncio.run(main())

Claude Code

claude mcp add --transport http c4c https://mcp.claudeforcompliance.com/mcp

Other clients (Cursor, Copilot, ChatGPT and more): C4C MCP setup.

2. Your code: call the C4C API

The same tools over plain HTTPS: POST https://mcp.claudeforcompliance.com/v1/tools/<tool> with JSON arguments and Authorization: Bearer <key>. The OpenAPI 3.1 file needs no key and works with code generators and Postman. Get a free trial key: sign up with your work email, confirm the link, and the key is yours in about a minute.

Check a quote before it reaches a memo

verify_quote is the step most homegrown tools skip. Hand it the passage your model produced; it says whether the regulator actually wrote it. A real line from 24 CFR 203.402 comes back exact. A paraphrase does not:

curl -s https://mcp.claudeforcompliance.com/v1/tools/verify_quote \
  -H "Authorization: Bearer $C4C_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Hazard insurance premiums on the mortgaged property at any rate",
       "register_id": "fha-24cfr-203-402"}'

"outcome": "not_found",
"unsupported": ["at any rate"],
"differences": {
  "extra_in_quote": ["at any rate"],
  "closest_source": "Hazard insurance premiums on the mortgaged property"
},
"note": "... A dropped or added word (for example 'not') can reverse the meaning."

Full tool list, Python and JavaScript examples, errors and limits: C4C API docs.

3. Your environment: run C4C Private

Want the library inside your own stack, next to your own data and your own retrieval? Run C4C Private. It is the whole library behind a small read-only server in your network or AWS account. It never sees your documents, makes no outside connections once updates come through your own gate, and every release is signed.

git clone https://github.com/earl-hash/c4c-mcp-server.git
cd c4c-mcp-server
docker build -t c4c-mcp .
docker run -d --name c4c-mcp -p 8080:8080 -v c4c-data:/data \
  -e C4C_LICENSE_KEY=your-free-key \
  -e C4C_AUTH_TOKEN=choose-a-long-random-token \
  c4c-mcp
# then point any MCP client at http://<host>:8080/mcp
# with the header  Authorization: Bearer <C4C_AUTH_TOKEN>

• License key: free with a work email. It only fetches signed library releases. Free keys get the free edition, refreshed a few times a year, for research and reference; paid update plans get every release and the right to make work product from it.
• AWS: a free AWS Marketplace listing for ECS and EKS, with a CloudFormation template. Update plans work the same on AWS.
• Azure: a free Microsoft Marketplace listing that deploys to Azure Container Apps in your own subscription.
• Details: C4C Private, and the install, data-flow and upgrade guides on GitHub.

What “formatted for AI” means here

• One section at a time, with its pinpoint citation: about 200 tokens on average, not a 1,000-page guide your model can’t fit in context.
• The regulator’s exact words, kept apart from any summary.
• Version and currency on every section: superseded, superseded in part, two versions in force, pending changes.
• Where it came from: the source URL and the date we retrieved it.
• A quote check (verify_quote), so a tool can prove its quotes before a person relies on them.

Research and reference only, not legal advice; every result says so. The code is public under the Elastic License 2.0; the library is free for research and reference use under the C4C Library License. API use is subject to the API terms. Questions or a bigger plan: earl@cfsynergies.com.