FDA MCP server · AI workflow integration

Connect your AI workflow to source-linked FDA medical-device evidence.

Stop pasting FDA PDFs into one-off chats. Add Constat MCP once, then let ChatGPT, Claude, Codex, Cursor, VS Code, or your own agent query the same structured 510(k), postmarket, reimbursement, and compliance evidence—with source boundaries attached.

Built for teams looking for an FDA medical-device API for AI agents, a Claude FDA connector, an OpenAI FDA integration, or a reusable regulatory-intelligence AI tool.

Direct answer

Yes. Constat can be connected now to any MCP-compatible agent or client that supports remote Streamable HTTP.

Public read tools require no authentication. Platform plans, administrator settings, approval policies, and tool filters still apply in the client you choose.

https://constat.dev/api/mcp
ChatGPTOpenAI APIClaudeClaude DesktopClaude CodeCodexCursorVS CodeGitHub CopilotCloudflare AgentsLangChainLlamaIndex

Put FDA evidence inside the work—not beside it

510(k) predicate research

Search devices like yours, inspect intended-use and validation fields, and trace predicate chains without rebuilding a one-off FDA scraper.

Example agent prompt

Find Radiology AI devices that report clinical data. Compare their validation designs and preserve every FDA source quotation.

Predicate-search method

AI-device evidence review

Retrieve the evidence FDA published for a specific K-number or De Novo, including reported sample sizes, endpoints, metrics, PCCPs, and source pages.

Example agent prompt

For K252148, summarize the reported validation evidence, identify what was not stated, and cite the source page for every extracted field.

Validation-evidence method

Postmarket and compliance monitoring

Bring recalls, MAUDE trends, warning-letter matches, Form 483 citations, and watchlist changes into an agent workflow with attribution limits intact.

Example agent prompt

Show the recent public postmarket signals for K230144 and separate device-attributable observations from product-code context.

MAUDE-search method

Clearance-to-payment research

Trace a clearance number into documented NTAP, CPT, HCPCS, CMS-rate, LCD, and public payer-policy evidence without pooling unlike payment measures.

Example agent prompt

Trace DEN170073 from FDA clearance to documented reimbursement mechanisms. Keep each amount, date, and source separate.

Reimbursement method

Connect where your team already works

Constat is a hosted MCP server, not another dashboard your agent must scrape. Choose a supported connection path, add one URL, review the tool catalog, and keep your existing model, IDE, orchestration framework, and approval policy.

Chat and research assistants

ChatGPT

Register the endpoint as a remote MCP connection in ChatGPT developer mode, then review and enable the imported tools.

Official connection documentation ↗

Claude and Claude Desktop

Open Settings → Connectors → Add custom connector and paste the Constat endpoint. Availability can depend on plan and workspace policy.

Official connection documentation ↗

Claude Code

Run: claude mcp add --transport http constat https://constat.dev/api/mcp

Official connection documentation ↗

OpenAI and coding agents

OpenAI Responses API

Add an MCP tool with server_url set to the Constat endpoint. Keep tool approval enabled for consequential workflows.

Official connection documentation ↗

OpenAI Codex

Add a Streamable HTTP server under mcp_servers.constat in Codex config. The public endpoint does not require a bearer token.

Official connection documentation ↗

Cursor

Add Constat as a remote URL in a project or user mcp.json file; Cursor supports Streamable HTTP MCP servers.

Official connection documentation ↗

IDE and developer workflows

VS Code and GitHub Copilot

Use MCP: Add Server or add an HTTP server entry to mcp.json, then make the tools available in agent chat.

Official connection documentation ↗

GitHub Copilot CLI

Run: copilot mcp add --transport http constat https://constat.dev/api/mcp

Official connection documentation ↗

Any compatible MCP client

Use the remote Streamable HTTP transport and the endpoint below. Client-specific approval and tool-filter settings still apply.

Official connection documentation ↗

Agent frameworks

Cloudflare Agents

Call addMcpServer with the Constat URL, discover its tools, and pass the selected tools into your agent model call.

Official connection documentation ↗

LangChain and LangGraph

Configure MultiServerMCPClient with transport http and the Constat URL, then pass the loaded tools to your agent.

Official connection documentation ↗

LlamaIndex

Create a BasicMCPClient with the Constat URL and convert the server tools into a tool specification for your agent.

Official connection documentation ↗

Copy a working start

Claude Code

claude mcp add --transport http constat https://constat.dev/api/mcp

Codex config.toml

[mcp_servers.constat]
url = "https://constat.dev/api/mcp"

OpenAI Responses API tool

{
  "type": "mcp",
  "server_label": "constat",
  "server_description": "Source-linked FDA medical-device evidence",
  "server_url": "https://constat.dev/api/mcp",
  "require_approval": "always"
}

Generic remote MCP client

{
  "mcpServers": {
    "constat": {
      "type": "http",
      "url": "https://constat.dev/api/mcp"
    }
  }
}

Need framework code instead? Use the maintained direct, LangChain, LlamaIndex, and Cloudflare examples.

Why use an FDA MCP server instead of a generic search prompt?

Typed research tools

Your agent selects a bounded lookup, search, chain, cohort, or monitoring operation instead of improvising a scraping plan.

Source-linked fields

Extracted evidence can include the verbatim source quotation and page so the primary record remains inspectable.

Honest missingness

Not reported, not found, unavailable, and not-yet-parsed are kept distinct rather than silently converted into a conclusion.

Workflow portability

Use the same endpoint across assistants, IDEs, and frameworks without rebuilding the FDA evidence layer for every model.

Keep the decision boundary human

  • Verify consequential findings against the linked FDA, CMS, or payer record.
  • Preserve each denominator and do not pool unlike performance or payment measures.
  • Do not treat missing public evidence as proof of absence, safety, or compliance.
  • Keep regulatory, clinical, legal, coding, coverage, and payment decisions under qualified review.
Review the architecture and trust boundaries →

Questions people ask about connecting FDA data to AI

What is an FDA MCP server?

An FDA MCP server exposes FDA-related tools to AI applications through the Model Context Protocol. Constat MCP provides typed tools for source-linked medical-device premarket evidence, postmarket signals, reimbursement pathways, and compliance records.

Can ChatGPT connect directly to Constat?

Yes. Where ChatGPT developer connections are available, register https://constat.dev/api/mcp as a remote MCP server, review the discovered tools, and enable the tools appropriate to the conversation. Constat is not claiming an accepted OpenAI directory listing unless that listing is separately verified.

Can Claude and Claude Desktop use Constat as a connector?

Yes. On supported Claude plans and workspaces, add https://constat.dev/api/mcp under Settings, Connectors, Add custom connector. Claude Code can add the same endpoint with its mcp add command.

Can I use Constat with an OpenAI API agent?

Yes. The OpenAI Responses API accepts public remote MCP servers through an MCP tool definition with a server_url. Set that URL to the Constat endpoint and choose an approval policy appropriate for your application.

Which AI agents and developer tools can connect?

Constat works with MCP clients that support remote Streamable HTTP. Verified connection paths include ChatGPT developer connections, the OpenAI Responses API, Claude, Claude Desktop, Claude Code, Codex, Cursor, VS Code with GitHub Copilot, GitHub Copilot CLI, Cloudflare Agents, LangChain, and LlamaIndex.

Does Constat replace FDA, regulatory, clinical, legal, or reimbursement review?

No. Constat organizes descriptive evidence from named public sources. It does not make clinical decisions, certify safety, predict enforcement, determine substantial equivalence, or guarantee coding, coverage, or payment. Verify consequential decisions against the primary record and qualified review.

Is the Constat MCP endpoint free to try?

Yes. The public read surface can be connected without authentication and is rate-limited. Higher-volume, licensed, or internally configured workflows should be scoped with Health AI before production use.

One endpoint. Your workflow.

Give your agent FDA evidence it can inspect, cite, and hand back for review.

Start with the public endpoint, or scope higher-volume access, configured monitoring, and evidence requirements for your regulatory, quality, market-access, or agent team.

Statistical pattern analysis of public FDA / CMS data — descriptive decision-support, not legal, regulatory, coding, or reimbursement advice. Verify against primary sources. A Health AI product · machine access over MCP · terms & dataset license.