How to Connect Libby Health to Claude With MCP

To connect Libby Health to Claude, add Libby's remote MCP server as a Claude connector, complete Libby OAuth, enable the connector for a conversation, and test it against a report already in your account. The current server URL, verified July 21, 2026, uses the https://app.libby.health host and /api/mcp/mcp path. Copy the complete URL from the live Libby MCP documentation while you connect because Claude menus, plan eligibility, and administrator controls can change.

The connector gives Claude structured, read-only access to the lab information available through nine Libby tools. It does not give Claude a write path into your record. That boundary matters, but it is not a privacy shield: when Claude invokes a tool, the requested structured result is sent to the connected AI service and becomes part of that provider interaction.

This is a record-organization workflow, not medical interpretation. Use Claude to retrieve or summarize source facts and prepare questions. Check important details against the original report, and keep diagnosis, treatment, medication, and urgency decisions with a qualified clinician.

What MCP changes in this workflow

The Model Context Protocol, or MCP, is an open standard for connecting AI applications to external systems. The official MCP introduction describes a client-server pattern in which an AI application can use tools or data exposed by an MCP server. In this case:

• Claude is the MCP-aware AI client.
• Libby operates the remote MCP server.
• OAuth 2.1 with PKCE authorizes the connection to one Libby account.
• Claude can invoke only the tools that Libby's server exposes.

That is different from uploading a PDF into one chat. An upload gives the model a copy of the selected file. The connector lets Claude request a structured result from the connected Libby account when a relevant tool is enabled and invoked. Neither path guarantees that the model's summary or explanation is correct.

Anthropic's current connector guide says connectors can be enabled per conversation and that Team and Enterprise organizations have owner controls. Its remote-custom-connector feature is still labeled beta, so treat the live guide—not screenshots or an old menu path—as the authority for Claude availability and UI.

What the Libby connector can—and cannot—return

Libby's current MCP documentation defines nine tools, all marked read-only and scoped to the OAuth-connected user:

• ping confirms that the authenticated connector is available.
• list_markers returns recorded markers with the latest value, unit, flag, data-point count, and an available lab reference range.
• get_marker_history returns time-ordered readings for one or more marker identifiers.
• search_markers resolves a familiar marker name to Libby's catalog and says whether the connected record has data.
• list_reports lists lab reports and panels, newest first, with dates and marker counts.
• get_report returns one report or panel with its structured marker results.
• check_allergen checks whether a named allergen IgE test is present and, if tested, returns the recorded sensitization result.
• list_allergen_sensitizations returns positive allergen IgE sensitizations and the total number of allergen IgE tests in the record.
• get_profile returns age in years and biological sex for reference-range context; it does not return raw date of birth.

The connector does not expose the uploaded document or image itself. It also does not return account credentials, name, email, street address, raw date of birth, internal database identifiers, or proprietary target bands. It has no create, update, upload, or delete tool.

Those are connector boundaries, not claims about what every Claude surface stores, retains, or uses. The record data returned by a tool crosses from Libby to the AI service you authorized. Review both services' current privacy and data controls before connecting personal health data. Read-only means Claude cannot change the Libby record through these tools; it does not mean the returned values remain inside Libby.

A connector can narrow what Claude may request. It does not make the returned answer a source record.

How to connect Libby Health to Claude

These steps were verified on July 21, 2026. Use the linked live documentation if the interface has moved.

Step 1: prepare the Libby record

Sign in to the Libby account you intend to authorize. Connector access currently requires full Libby access through an active trial or paid membership. Make sure at least one lab report has finished processing; the connector can return only structured information already present in that account.

If the underlying record still needs work, first follow the source-preserving workflow for organizing years of blood test results. The original report remains the check against any extracted or AI-generated summary.

Step 2: copy the current remote MCP server URL

Open the Libby MCP connector documentation and copy the server URL shown there. As of this review, it is:

https://app.libby.health + /api/mcp/mcp

Use that exact HTTPS endpoint. Do not substitute the marketing homepage, an app dashboard URL, or an MCP endpoint copied from an unofficial directory.

Step 3: add Libby as a custom connector in Claude

In Claude, open the current connector-management surface and choose Add custom connector. Enter Libby Health as the name and paste the remote MCP server URL.

Anthropic currently directs individual users through Customize → Connectors. On Team and Enterprise plans, an Owner or Primary Owner may need to add the custom connector under Organization settings → Connectors before a member can connect. Directory listing is not required: if Libby is not listed, use the custom remote connector path.

Step 4: complete OAuth and enable the connector

Choose Add or Connect, then complete the Libby sign-in and OAuth consent flow. Confirm that the account shown is the Libby account whose record you mean to use. Review the requested permissions rather than treating the consent screen as a routine click-through.

After authorization, enable Libby for the conversation where you want to use it. Claude's current guide places conversation-level connectors behind the plus menu near the composer. Organization policy or client availability can change what you see.

Step 5: run a narrow test and verify the result

Start with one fact that is easy to check against Libby, such as:

• "List the reports in my connected Libby account."
• "Show the latest recorded HbA1c value, unit, date, flag, and lab reference range."
• "Show the recorded history for ApoB, including dates and units."

Then compare the response with the Libby record and original report. Check the test name, value, unit, collection date, flag, and report-specific range. If a range is absent, the connector should return that absence rather than invent one. The guide to understanding lab reference ranges explains why a range comparison is still not a diagnosis or personal target.

Prompts that fit the connector's real scope

Good first prompts map cleanly to one or more documented tools:

• "Which lab markers are available in my Libby record?"
• "Show my recorded fasting glucose history with dates and units."
• "Summarize my newest lab panel and keep values, units, flags, and lab ranges together."
• "Was I tested for peanut IgE, and what recorded result is available?"
• "List the positive allergen IgE sensitizations recorded in my account."

Ask Claude to name missing data and uncertainty. Do not ask it to fill gaps, invent a reference range, decide that differently named tests are comparable, or turn an out-of-range flag into a diagnosis. For a broader audit workflow, use how to use lab results with ChatGPT or Claude.

Troubleshooting the connection

The custom connector option is missing

Check Anthropic's live custom-connector documentation for the Claude product, plan, and organization you are using. Custom remote connectors are a beta feature, and an organization owner may need to enable or add the connector. Anthropic's current guide says custom remote servers are added through Claude web or Desktop settings. Mobile can use a remote connector that was already added, but it cannot add one directly.

OAuth loops or the connection fails

Confirm that you are signed in to the intended Libby account in the same browser and that the account has full access. Remove the failed connector from Claude, add the current endpoint again, and repeat OAuth. If a managed organization is involved, ask its owner whether custom connectors are allowed.

Claude connects but returns no markers or reports

Open Libby and confirm that at least one report has finished processing in the same account used for OAuth. A successful connector handshake proves the server is reachable; it does not prove the connected record contains data.

A marker or report is not found

Ask Claude to list available markers or reports first. For a marker, try a familiar name so search_markers can resolve it. A missing result may mean the report is not processed, the connected account has no such result, or the query name did not resolve. Do not treat absence from the connector as evidence that a test was never performed.

Claude answers without using Libby

Confirm that Libby is enabled for that conversation, then ask a question tied to a specific documented tool. Make the source explicit: "Use the Libby Health connector to list my available reports." Check the tool-use details before assuming an answer came from your record.

A returned value looks wrong

Stop and compare the value, unit, date, and range with the original report. Do not make a health decision from the disputed output. When contacting Libby support, include the connector name, failed step, and visible error message—but do not email lab values or medical records.

The privacy and safety boundary to remember

Anthropic's remote connector security guidance recommends connecting only to trusted servers, reviewing OAuth permissions, watching tool inputs and outputs, and considering prompt-injection risk. It also explains that remote connectors are reached from Anthropic's cloud, including when configured through Claude Desktop.

For Libby specifically, the server is first-party and the nine current tools are read-only. That narrows what the connector can do to the Libby record. It does not answer separate questions about chat retention, model training, human access, deletion, or organization policy at the AI provider. Review the current Claude privacy controls and Libby app privacy policy before authorizing the connection. The checklist for sharing health records with an AI service applies to a connector too: identify the product surface, minimize the task, keep the source, verify the output, and disconnect access you no longer need.

FAQ

Is Libby Health listed in Claude's connector directory? Do not assume it is. If Libby Health is not listed, use Claude's Add custom connector path and the server URL in Libby's live MCP documentation. Directory listings and menus can change.

What can Claude read through the Libby connector? The nine documented tools can return marker summaries and histories, report or panel contents, allergen IgE results, and age in years plus biological sex for range context. Results are limited to the OAuth-connected Libby account.

Can Claude change my Libby record? No. Libby's current MCP server exposes no create, update, upload, or delete tool. Read-only describes the connector's effect on Libby; it does not describe every way the AI provider may handle returned data.

Does read-only mean my health data stays inside Libby? No. When Claude invokes an authorized tool, Libby returns structured record data to the connected AI service. Review both services' current privacy, retention, training, and deletion controls before connecting.

Can Claude see the PDFs or images I uploaded to Libby? Not through the current MCP tools. The connector returns structured results and report data, but Libby's documentation says it does not return uploaded documents or images.

Why does Claude show no markers after I connect? Check that a lab report has finished processing in the OAuth-connected Libby account and that the account has full access. Then ask Claude to list available reports or markers before requesting a specific one.

Can I connect Libby to ChatGPT with the same MCP server? Libby's live MCP documentation includes a ChatGPT path using the same server and Libby OAuth. OpenAI's current Connect from ChatGPT guide says apps have been supported on all plans since November 13, 2025. Separately, it says an organization can control whether Developer mode is available, so follow the current client-specific guide rather than assuming the toggle is enabled.

Can Claude interpret the results or tell me what treatment to choose? Claude can retrieve, organize, or summarize recorded information, but neither the connector nor an AI answer establishes a diagnosis or treatment plan. Verify source facts and discuss medical meaning and decisions with a qualified clinician.

References

• Model Context Protocol: What Is MCP?
• Anthropic: Use Connectors to Extend Claude's Capabilities
• Anthropic: Get Started With Custom Connectors Using Remote MCP
• OpenAI: Connect From ChatGPT
• Anthropic Privacy Center
• Libby Health App Privacy Policy