Connect the FDA record to your agent.
Connect once, start with a company-level manufacturing and compliance summary, then drill into inspections, warning letters, OII records, facilities, import risk, and product data as needed.
Jump to section
Start here
Pick the app you use
Each link opens the setup steps for that app. What the server covers is on the FDA MCP server page.
Ask a first question
Try one of the prompts below, with a company you already know.
Evaluating a company for manufacturing, compliance, diligence, or account research? Start with fda_manufacturing_risk_summary, then follow the trail.
- Start with
fda_manufacturing_risk_summarywhen you need a fast company-level view. - Use
fda_facility_dossierwhen you already know the FEI and want the local compliance story fast. - Use
fda_company_compliance_timelinewhen you want one dated event stream across inspections, warning letters, import risk, and recalls. - Use inspections, citations, warning letters, OII, import alerts, and debarments after the workflow tools when you need raw evidence.
- Use product, label, approval, shortage, and recall tools after you know which business unit or facility you care about.
“Give me a manufacturing risk summary for Pfizer.”
“Summarize recent warning letters, inspections, and import alerts for Moderna.”
“Show manufacturing risk summary plus recent OII records for Thermo Fisher.”
Authentication
Provision a key on /signup. There are two ways to use it:
- OAuth sign-in (claude.ai and ChatGPT connectors). Add the plain URL
https://www.regdatalab.com/mcpas a custom connector and keep the sign-in settings the client detects. RegDataLab opens a sign-in page: enter your email and open the link we send in the same browser. No account yet? Approving the connection creates a free one. Pasting an API key works too. The server then issues the connector its own tokens, so a key never goes in the URL. - Bearer header (everything else). Claude Desktop, Claude Code, Cursor, JSON configs, and raw HTTP clients send the key in the
Authorizationorx-api-keyheader. Persist the session id returned byinitializefor the life of the connection. A client that can only take a URL signs in with OAuth athttps://www.regdatalab.com/mcp. The server refuses a key placed in the URL query string with HTTP 400.
The key on the homepage is a shared demo key. It is not a personal key: every visitor uses the same credential, it is limited to 10 requests per minute and 50 credits per IP address, and it can only call the public FDA tools. Sign up for your own key to keep credits and account access private.
Authorization: Bearer fda_...
# or: x-api-key: fda_...
# Persist the session id returned by initialize:
Mcp-Session-Id: <session_id>Connect your client
RegDataLab speaks the Model Context Protocol over one HTTP endpoint. Start with claude.ai or ChatGPT for a zero-install connection, or wire it into Claude Desktop, Claude Code, Cursor, VS Code, Codex, the OpenAI API, or your own pipeline. Every one of those signs in, so no key has to go into a config file. In the key-based examples, replace YOUR_API_KEY with the key from /signup.
No install
Other clients
C
Claude Desktop
Add the connector once in claude.ai and it appears in Claude Desktop and Cowork, with nothing to install. The local wrapper stays available for a stdio-only setup.
- 01Add it once in claude.ai (open claude.ai/directory/regdatalab and click Connect) — connectors you add there appear in Claude Desktop and Cowork
- 02Local alternative, if you would rather not use the connector: Settings → Developer → Edit Config, paste the config below, then quit Claude completely and reopen it (needs Node.js)
https://www.regdatalab.com/mcpLocal alternative (needs Node.js): Settings → Developer → Edit Config, paste this into claude_desktop_config.json, then quit Claude completely and reopen it. The file paths are ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and %APPDATA%\Claude\claude_desktop_config.json on Windows.
{
"mcpServers": {
"regdatalab": {
"command": "npx",
"args": ["-y", "fda-data-mcp"],
"env": { "FDA_DATA_API_KEY": "YOUR_API_KEY" }
}
}
}D
Claude Code
One command, native HTTP transport, no key to paste: Claude Code signs in and keeps its own token.
- 01Run the command below once, then in Claude Code run /mcp, pick regdatalab and choose Authenticate
- 02Already connected in claude.ai? It shows up in /mcp on its own, with nothing to install
claude mcp add --transport http --scope user regdatalab https://www.regdatalab.com/mcpFor CI or a machine with no browser, pass the key as a header instead:
claude mcp add regdatalab https://www.regdatalab.com/mcp \
--transport http \
--header "Authorization: Bearer YOUR_API_KEY" \
--scope userE
Cursor
One click to install, then sign in. Nothing is stored in the link, and no key lands in your config file.
- 01Click Add to Cursor, then Install (on a phone, open this page on your computer instead)
- 02In Cursor, open Settings → MCP → regdatalab and click Connect, then approve on the RegDataLab page
- 03No deeplink? Put the config below in ~/.cursor/mcp.json and click Connect
{
"mcpServers": {
"regdatalab": {
"url": "https://www.regdatalab.com/mcp"
}
}
}API key instead of signing in (Cursor expands ${env:VAR}, so the key stays in your environment):
{
"mcpServers": {
"regdatalab": {
"url": "https://www.regdatalab.com/mcp",
"headers": { "Authorization": "Bearer ${env:REGDATALAB_API_KEY}" }
}
}
}F
VS Code (GitHub Copilot)
Native HTTP transport with sign-in. VS Code's config key is servers, not mcpServers.
- 01Command Palette → MCP: Add Server → HTTP, paste this URL and name it regdatalab (or use the one-click install link below)
- 02Start the server when VS Code offers to, and approve the sign-in page that opens
- 03Project-wide instead: put the config below in .vscode/mcp.json
{
"servers": {
"regdatalab": {
"type": "http",
"url": "https://www.regdatalab.com/mcp"
}
}
}G
Codex CLI and ChatGPT desktop
The Codex CLI, the Codex IDE extension and ChatGPT desktop share ~/.codex/config.toml.
- 01Run both commands below: the first adds the server, the second signs you in
- 02ChatGPT desktop instead: Settings → MCP servers → Add server → Streamable HTTP, paste this URL, save and authenticate
codex mcp add regdatalab --url https://www.regdatalab.com/mcp
codex mcp login regdatalabOr write it straight into ~/.codex/config.toml:
[mcp_servers.regdatalab]
url = "https://www.regdatalab.com/mcp"
# API key instead of signing in:
# bearer_token_env_var = "REGDATALAB_API_KEY"H
OpenAI Responses API
Pass RegDataLab as an MCP tool in the OpenAI Responses API, with your key in the Authorization header.
from openai import OpenAI
client = OpenAI()
resp = client.responses.create(
model="gpt-4.1",
tools=[{
"type": "mcp",
"server_label": "regdatalab",
"server_url": "https://www.regdatalab.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}],
input="Look up Pfizer's FDA facilities",
)I
Your own code (plain HTTP)
Any HTTP client works: the endpoint takes MCP JSON-RPC requests with your key in the Authorization header. Open a session with initialize, send the Mcp-Session-Id it returns on every later request, then call tools/list and tools/call. Replies arrive as server-sent events; the JSON result is on the data: line.
KEY=YOUR_API_KEY
URL=https://www.regdatalab.com/mcp
H=(-H "Authorization: Bearer $KEY" -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream")
# 1. Open a session. The id comes back in the Mcp-Session-Id header.
SID=$(curl -s -D - -o /dev/null "$URL" "${H[@]}" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}' \
| awk 'tolower($1)=="mcp-session-id:"{print $2}' | tr -d '\r')
curl -s "$URL" "${H[@]}" -H "Mcp-Session-Id: $SID" \
-d '{"jsonrpc":"2.0","method":"notifications/initialized"}'
# 2. List the tools.
curl -s "$URL" "${H[@]}" -H "Mcp-Session-Id: $SID" \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'
# 3. Call a tool (1 credit).
curl -s "$URL" "${H[@]}" -H "Mcp-Session-Id: $SID" \
-d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"fda_resolve_company","arguments":{"company":"Pfizer"}}}'J
Team sharing (Claude Code .mcp.json)
Drop this in your project root for team-wide access in Claude Code, with the key resolved from an environment variable. Cursor and VS Code use their own files: see the Cursor and VS Code tabs.
{
"mcpServers": {
"regdatalab": {
"type": "http",
"url": "https://www.regdatalab.com/mcp",
"headers": {
"Authorization": "Bearer ${REGDATALAB_API_KEY}"
}
}
}
}A
Connect in claude.ai (no install)
The fastest path — nothing to install, works in any modern browser.
- 01Open claude.ai/directory/regdatalab and click Connect (Team and Enterprise: an owner adds it in Organization settings → Connectors first, then you click Connect)
- 02On the RegDataLab page that opens, enter your email and open the sign-in link we send in the same browser (no account or API key needed; pasting a key works too), then approve
- 03In a chat, turn RegDataLab on under + → Connectors
- 04Listing not shown for you? Customize → Connectors → Add custom connector, name it RegDataLab and paste this URL instead
https://www.regdatalab.com/mcpFree plans can add one custom connector.
B
Connect in ChatGPT (no install)
ChatGPT creates a custom MCP server as a plugin. You add it on chatgpt.com in a browser, and your plan or workspace must allow custom MCP servers.
FDA company records in ChatGPT
- 01On chatgpt.com, open Plugins, click Add and choose Add custom MCP server (your plan or workspace must allow it)
- 02Name it RegDataLab, paste this URL as the Server URL, set Authentication to OAuth, confirm the notice and click Create as a plugin
- 03On the RegDataLab page that opens, enter your email and open the sign-in link we send in the same browser (no account or API key needed; pasting a key works too), then approve
- 04Install the plugin, then in a chat type @ and pick RegDataLab
https://www.regdatalab.com/mcpTroubleshooting
- Tools not showing after setup
- Restart the client after editing its config, then confirm regdatalab appears in the connector or MCP server list. In Claude, turn the connector on in the chat under + → Connectors. In ChatGPT, install the plugin and type @RegDataLab in the chat.
- Invalid API key or “This API key is not recognized”
- The key was replaced (for example by key recovery), its account was deleted, or it was copied incompletely. Get your current key from /account; if you use a hosted MCP service, update the key saved there too. Confirm the Bearer token is the full key from /signup with no spaces or line breaks. For claude.ai and ChatGPT connectors, disconnect and sign in again, then paste the full key on the RegDataLab page.
- Claude.ai detected “No sign-in”
- Remove the connector and add it again with the exact URL https://www.regdatalab.com/mcp, then choose Sign in now as the authentication. On the RegDataLab page, enter your email and open the sign-in link we send, or paste your API key.
- npx or fda-data-mcp not found
- Install Node.js, then test the wrapper directly: FDA_DATA_API_KEY=YOUR_API_KEY npx -y fda-data-mcp. Wrapper documentation: https://github.com/medley/fda-data-mcp
- Session expired
- Re-initialize by restarting the client. Most clients handle session renewal automatically.
Search tips
- Use the most canonical company name you know first, then confirm aliases or subsidiaries before widening the search.
- For large parents, drill down into a specific subsidiary, brand, FEI, or geography when broad name matching brings back noise.
- Avoid firing many heavy tools in parallel until you know which company entity and question you actually want to pursue.
- Use fda_resolve_company to jump straight to the right corporate family and its sites.
- Credits are consumed per tool call that runs, including empty or not-found results.
- Calls rejected for invalid or missing input, and calls that fail with an internal server error (code internal_error), are refunded.
- When credits are exhausted the API returns HTTP 402 with a no_credits-style error instead of silently failing.
- New keys start with 50 credits at signup; verify your email to unlock 300 credits a month, at 60 requests per minute.
Who it helps
Built first for manufacturing and supplier-quality questions about companies and facilities. The same records also support the workflows below.
Tool reference
51 MCP tools, organized by category. Claude and ChatGPT connectors list all 51, as do API-key sessions. Start with the manufacturing summary for a fast company view, then reach for the drill-down tools when you need raw evidence. Each row shows the credits one call uses (also listed on Pricing).
02Start Here
| Tool | Description | Credits |
|---|---|---|
fda_manufacturing_risk_summary | Best first tool for a company-level manufacturing and compliance summary across inspections, warning letters, OII records, import risk, debarments, facilities, and recalls. | 3 credits |
fda_compare_companies | Compare 2 to 5 companies side by side in one call: facilities, inspection classifications, warning letters, the latest Form 483, import alerts and refusals, debarments and recalls, each with its most recent date. | 6 credits |
14Manufacturing & Compliance
| Tool | Description | Credits |
|---|---|---|
fda_inspections | Search FDA inspections by company, FEI, date, and geography. NAI means no objectionable conditions or practices were found during the inspection. VAI means objectionable conditions or practices were found, but the agency is not prepared to take or recommend any administrative or regulatory action. OAI means regulatory and/or administrative actions are recommended. | 1 credit |
fda_inspection_stats | Count FDA inspections by fiscal year and outcome (OAI, VAI, NAI) with the OAI share, filtered by product type, project area or country. | 1 credit |
fda_citations | Search inspection citations and CFR findings tied to facilities and companies. | 1 credit |
fda_compliance_actions | Search warning letters, seizures, and injunctions from FDA compliance dashboards. | 1 credit |
fda_search_warning_letters | Search FDA warning letters with dates, issuing office context, and text or excerpts where captured. | 1 credit |
fda_search_oii_records | Search OII reading-room records, including machine-readable inspection and compliance documents. | 1 credit |
fda_search_debarments | Search FDA debarment records and integrity-related exclusions. | 1 credit |
fda_search_import_alerts | Search import alerts, red-list and green-list placements, and linked firms. | 1 credit |
fda_search_opdp_untitled_letters | Search OPDP untitled letters for pharma promotional and commercial compliance issues. | 1 credit |
fda_bpdr_summary | Review BPDR annual summary data for biologics manufacturing and quality signal context. | 1 credit |
fda_device_supply_status | Check device supply disruption and shortage signals for medtech manufacturing monitoring. | 1 credit |
fda_inspection_observation_summary | Review annual inspection observation summary trends and common citation patterns. | 1 credit |
fda_import_refusals | Search import refusal records by company, FEI, country, category, and date. | 1 credit |
fda_ires_enforcement | Search iRES enforcement and recall records from FDA internet Recall Enterprise System data. | 1 credit |
10Company & Facility Intelligence
| Tool | Description | Credits |
|---|---|---|
fda_lookup_company | Quick company view across facilities, enforcement history, and alias resolution. | 1 credit |
fda_search_facilities | Search FDA facilities by company, location, and operating context. | 1 credit |
fda_get_facility | Open a detailed facility record by FEI, including linked enforcement context. | 1 credit |
fda_facility_products | List products and classifications linked to a facility. | 1 credit |
fda_resolve_company | Resolve a company to its normalized aliases and canonical record. | 1 credit |
fda_suggest_subsidiaries | Suggest likely subsidiaries and brand/entity relationships from FDA and EDGAR-style evidence. | 2 credits |
fda_company_full | Pull a broader company profile across facilities, recalls, approvals, and drug/device coverage. | 5 credits |
fda_company_compliance_timeline | One reverse-chronological event stream across inspections, warning letters, import risk, debarments, and recalls. | 2 credits |
fda_facility_dossier | Full compliance dossier for one facility by FEI: inspections, citations, warning letters, recalls, and import risk in a single call. | 1 credit |
fda_search_family_facilities | Map a company's entire manufacturing footprint across its resolved corporate family. | 1 credit |
12Products, Approvals & Labels
| Tool | Description | Credits |
|---|---|---|
fda_search_510k | Search FDA 510(k) clearances by company, product code, and decision details. | 1 credit |
fda_search_pma | Search FDA PMA approvals by company, product code, and decision window. | 1 credit |
fda_search_drugs | Search Drugs@FDA applications, brand names, and submission status. | 1 credit |
fda_search_ndc | Search the NDC directory by labeler, brand, or application number. | 1 credit |
fda_search_nsde | Search NDC SPL Data Element (NSDE) records by name, package NDC, and application number. A listing does not indicate FDA approval. | 1 credit |
fda_search_by_product | Search across drug and device product-oriented datasets by product name. | 2 credits |
fda_device_class | Look up device classification details by product code. | 1 credit |
fda_product_code_lookup | Cross-reference a product code across classification, 510(k), and PMA data. | 1 credit |
fda_clinical_result_letters | Search Complete Response Letters and related FDA transparency records. | 1 credit |
fda_drug_labels | Search SPL drug labels, sections, manufacturers, and application references. | 1 credit |
fda_device_udi | Search the FDA UDI dataset for device identifiers, companies, and product codes. | 1 credit |
fda_device_udi_lookup | Look up GUDID-style UDI records directly. | 1 credit |
08Recalls, Safety & Specialty
| Tool | Description | Credits |
|---|---|---|
fda_search_enforcement | Search recall and enforcement actions across drug, device, and food records. | 1 credit |
fda_search_recall_text | Search recall reason text and product descriptions across enforcement records. | 1 credit |
fda_recall_facility_trace | Trace recalls back to likely facilities and manufacturing entities. | 2 credits |
fda_drug_shortages | Search active and resolved FDA drug shortages. | 1 credit |
fda_device_recalls | Search device recall records with firm, product code, and status filters. | 1 credit |
fda_adverse_events | Search consumer (food, cosmetic), veterinary and tobacco adverse event reports. | 1 credit |
fda_substance_lookup | Look up FDA substance and UNII data by code or name. | 1 credit |
fda_search_aphis | Search USDA APHIS veterinary biologics establishments. USDA records, not FDA data. | 1 credit |
05Software Research (MES, QMS, LIMS)
| Tool | Description | Credits |
|---|---|---|
software_search_products | Search the public-source MES, QMS, LIMS and adjacent software dossiers by text and category. | 1 credit |
software_get_product | Profile summary, pricing stance, review status and key claims with sources for one product. | 1 credit |
software_compare | Compare 2 to 5 products side by side on deployment, pricing stance and regulatory posture. | 1 credit |
software_list_evidence | List a product's sourced claims by topic or source type, each with its URL and confidence. | 1 credit |
software_get_sources | Read a product's sources note: each source and what it supports. | 1 credit |
Rate limits
Rate limits are enforced per API key. Each key defaults to 60 requests per minute. The shared homepage demo key is not a personal key and is capped at 10 requests per minute and 50 credits per IP address, with 60 requests per minute shared by all visitors. When a limit is exceeded, the API returns HTTP 429 with the reset window in the response headers.
X-RateLimit-Limit: 60
X-RateLimit-Remaining: 42
X-RateLimit-Reset: 1708905600Errors
Errors use standard HTTP codes with JSON payloads.
| Code | Description |
|---|---|
401 / 403 | Authentication failure — invalid or missing API key. |
402 | Credits exhausted (REST API) — verify your email if you have not, upgrade to a paid plan on the pricing page, or wait for the monthly reset. MCP clients stay connected and the tool call itself reports the same thing. |
429 | Rate limit exceeded — retry after the reset window. |
500 | Internal error — retry or contact support. |
Data sources
RegDataLab ingests openFDA bulk downloads, DECRS facility data, FDA Dashboard compliance feeds, warning letters, OII reading-room records, import alerts, debarments, APHIS veterinary biologics data, and iRES enforcement reports. Each dataset is normalized with provenance metadata including source, ingested_at, and run_id.