Tebra

tebra-mcp-server

[npm version] [License: MIT]

MCP server for Tebra (formerly Kareo) practice management. Connects your existing Tebra account to Claude and other MCP-compatible AI agents, exposing 34 SOAP tools and 13 FHIR clinical tools for patients, encounters, appointments, billing, documents, insurance, and clinical data. Every request body is generated from the live Tebra WSDL contract (member names and sequence order verified against KareoServices.svc?xsd=xsd0/xsd7), with a regression suite locking the wire format in place. No data is accessible without valid Tebra API credentials.

This is an independent open-source project. It is not affiliated with, endorsed by, or supported by Tebra Technologies, Inc.; "Tebra" and "Kareo" are trademarks of their owner, used here only to identify the API this server talks to. Tool results contain protected health information: sending them to Claude requires a Claude plan under which Anthropic signs a Business Associate Agreement (BAA). Consumer Claude plans are not covered, and confirming coverage is the responsibility of the practice operating the server.

Hosted version available

Do not want to manage credentials, hosting, and updates yourself? DOCK is the managed version of this server: encrypted per-practice auth, audit logs, draft-first write actions, and a workflow library. Founding practices lock lifetime pricing: Front Desk $49/mo (Zenoti), Billing Desk $99/mo (Tebra; DOCK signs its own BAA with the practice, separate from Anthropic's), Full Practice $129/mo (both). https://dockhq.vercel.app

Quick Start

npx tebra-mcp-server

Prerequisites

• Node.js 18+
• Tebra SOAP API credentials (generated in Tebra PM admin under Settings > API)
• (Optional) Tebra FHIR API credentials for clinical data access

Environment Variables

SOAP API (required)

FHIR API (optional -- enables 13 clinical data tools)

FHIR credentials are obtained through Tebra appSphere. The server uses the OAuth2 client credentials flow with automatic token caching, refresh 60s before expiry, and a one-shot retry on 401. Two client authentications are supported: SMART Backend Services private_key_jwt (an appSphere Backend App registered as "Confidential (URL for JWKS)" — the server signs an RS384 client assertion with iss=sub=client id, aud=token URL, a 5-minute exp and a unique jti, sent as client_assertion with client_assertion_type=urn:ietf:params:oauth:client-assertion-type:jwt-bearer; the private key is read from disk once and never transmitted) and the legacy client_secret flow. Set the key path for the former, the secret for the latter. Note: both the practice and the backend-service client must be activated by Tebra Customer Care before tokens are issued — a 401 can mean "not yet activated" rather than "bad credentials".

Installation

Claude Code

Add to .mcp.json in your project root:

{  "mcpServers": {    "tebra": {      "command": "npx",      "args": ["-y", "tebra-mcp-server"],      "env": {        "TEBRA_SOAP_USER": "[email protected]",        "TEBRA_SOAP_PASSWORD": "your-password",        "TEBRA_CUSTOMER_KEY": "your-customer-key",        "TEBRA_FHIR_CLIENT_ID": "optional-fhir-client-id",        "TEBRA_FHIR_CLIENT_SECRET": "optional-fhir-client-secret"      }    }  }}

Claude Desktop

Add to your claude_desktop_config.json:

{  "mcpServers": {    "tebra": {      "command": "npx",      "args": ["-y", "tebra-mcp-server"],      "env": {        "TEBRA_SOAP_USER": "[email protected]",        "TEBRA_SOAP_PASSWORD": "your-password",        "TEBRA_CUSTOMER_KEY": "your-customer-key"      }    }  }}

Cursor / VS Code

Add to your MCP settings:

{  "mcpServers": {    "tebra": {      "command": "npx",      "args": ["-y", "tebra-mcp-server"],      "env": {        "TEBRA_SOAP_USER": "[email protected]",        "TEBRA_SOAP_PASSWORD": "your-password",        "TEBRA_CUSTOMER_KEY": "your-customer-key"      }    }  }}

Available Tools (47 total)

Patient Management

Appointments

Encounters & Billing

Insurance & Authorizations

Practice Configuration

Documents

Financial Analysis

External Vendor & System

FHIR Clinical Data (requires FHIR credentials)

These tools access clinical data via the Tebra FHIR R4 API. They require separate FHIR credentials (see Environment Variables above). If FHIR credentials are not configured, these tools will not be registered.

FHIR patient IDs are a different identifier space from SOAP patient IDs — use tebra_fhir_search_patients to resolve them. All FHIR searches follow Bundle pagination automatically (up to 10 pages), and the three resources where Tebra requires a second search parameter (MedicationRequest intent, CarePlan category, CareTeam status) get sensible defaults so they don't silently return empty.

Rate Limits

The SOAP client enforces a minimum interval between calls per action, mirroring the throttling thresholds in the Tebra API Technical Guide. When a tool is called more frequently than its limit allows, the client sleeps just long enough to satisfy the interval before sending the request — calls are delayed, never dropped.

On top of client-side throttling, every SOAP call retries up to 3 times with exponential backoff (1s, 2s, 4s) before surfacing an error. Use tebra_get_throttles to query Tebra's server-side rate limit counters in real time.

Example Workflows

Scheduling Flow

1. tebra_get_providers          -- Get provider IDs2. tebra_get_service_locations  -- Get location IDs3. tebra_get_appointment_reasons -- Get reason/type IDs4. tebra_create_appointment     -- Create with provider, location, reason IDs5. tebra_get_appointment_detail -- Verify creation

Encounter Approval Flow

1. tebra_create_encounter            -- Create superbill (status: Draft)2. tebra_update_encounter_status     -- Move to Submitted (shows as "Review" in Tebra's UI)3. tebra_update_encounter_status     -- Move to Approved (triggers billing)   OR3. tebra_update_encounter_status     -- Reject back to Draft

Front-Desk Check-In Flow

1. tebra_get_appointments             -- Today's schedule (resourceName = provider)2. tebra_update_appointment_status    -- CheckedIn on arrival3. tebra_update_appointment_status    -- CheckedOut at departure

Payment Posting Flow

1. tebra_search_patients    -- Find patient2. tebra_get_charges        -- Find outstanding charges3. tebra_create_payment     -- Post payment to patient account4. tebra_get_payments       -- Verify payment posted

Patient Onboarding

1. tebra_search_patients              -- Check for existing patient2. tebra_create_patient               -- Create if not found3. tebra_update_patient_external_id   -- Link Supabase client ID4. tebra_create_appointment           -- Schedule first visit

Clinical Context for Note Creation

1. tebra_get_appointments             -- Get today's schedule2. tebra_get_appointment_detail       -- Get appointment context3. tebra_get_patient                  -- Full patient demographics4. tebra_get_patient_authorizations   -- Check auth status5. tebra_fhir_get_allergies           -- Allergies6. tebra_fhir_get_medications         -- Current medications7. tebra_fhir_get_conditions          -- Problem list8. tebra_fhir_get_vitals              -- Recent vitals

Tool Dependency Chains

Some tools require IDs obtained from other tools. Key dependencies:

tebra_create_appointment  requires: patientId    (from tebra_search_patients or tebra_create_patient)  requires: providerId   (from tebra_get_providers)  requires: locationId   (from tebra_get_service_locations)  optional: reasonId     (from tebra_get_appointment_reasons)
tebra_create_encounter  requires: patientId    (from tebra_search_patients)  requires: providerId   (from tebra_get_providers)  recommended: practiceName/practiceId (from tebra_get_practices)  optional: authorization number (from tebra_get_patient_authorizations)
tebra_create_payment  requires: patientId    (from tebra_search_patients)
tebra_update_encounter_status  requires: encounterId  (from tebra_create_encounter or tebra_get_encounter)
tebra_create_document  requires: patientId    (from tebra_search_patients)
tebra_update_patient_external_id  requires: patientId        (from tebra_search_patients or tebra_create_patient)  recommended: externalVendorId (from tebra_get_external_vendors)
tebra_set_primary_patient_case  requires: patientCaseId (from tebra_get_patient — cases[].caseId)
All clinical tebra_fhir_get_* tools  require: FHIR patientId (from tebra_fhir_search_patients — NOT the SOAP patient ID)

Privacy Policy

This server runs locally and talks only to Tebra's API with your own credentials. It has no backend, no telemetry, and writes nothing to disk. Credentials and patient identifiers are scrubbed from the debug log and from error messages, and every record-returning tool takes a fields argument for minimum-necessary selection. Tool results contain protected health information, so sending them to Claude requires a Claude plan under which Anthropic signs a Business Associate Agreement. The full policy is in PRIVACY.md.

Desktop Extension (.mcpb)

manifest.json packages the server as an MCP Bundle for Claude Desktop, with every credential declared as sensitive user_config (stored in the OS keychain) and the FHIR private key as a file input. Build the bundle with:

npm run pack:mcpb   # builds, prunes dev dependencies, writes tebra-mcp-server.mcpb

src/__tests__/manifest.test.ts pins the manifest to package.json and to the tool registry, so a tool added to the server without a manifest entry fails the suite.

When installed in Claude Desktop, the FHIR private key must be an absolute path the desktop app can read (a Windows path such as C:\Users\you\keys\fhir.pem, not a path relative to some project). An unreadable key disables the 13 FHIR tools with a message on stderr; it does not stop the 34 SOAP tools. Optional fields left blank in the install form are tolerated: Claude Desktop passes them through as the literal ${user_config.<key>} placeholder, which the server treats as unset.

API Reference

The server wraps two Tebra APIs:

SOAP API v2.1 (34 tools)

• Endpoint: https://webservice.kareo.com/services/soap/2.1/KareoServices.svc
• Auth: RequestHeader with CustomerKey, Password, User (WSDL sequence order matters)
• Request bodies generated in WSDL (?xsd=xsd0/xsd7) member order — WCF silently drops out-of-order members
• All requests include retry with exponential backoff (3 attempts at 1s, 2s, 4s)

FHIR R4 API (13 tools)

• Endpoint: https://fhir.prd.cloud.tebra.com/fhir-request (note the hyphen — configurable via TEBRA_FHIR_BASE_URL)
• Auth: OAuth2 client credentials flow against https://fhir.prd.cloud.tebra.com/smartauth/oauth/token
• Token caching with automatic refresh before expiry and one-shot 401 retry
• Bundle pagination followed automatically (up to 10 pages per search)

Development

git clone https://github.com/jamesrosing/tebra-mcp-server.gitcd tebra-mcp-servernpm install
npm run dev    # tsx — runs src/index.ts directly without a build stepnpm run build  # tsc — compiles to dist/npm test       # node:test via tsx — 37 regression tests covering wire-format invariantsnpm start      # node dist/index.js — runs the compiled output

The regression suite pins the three Tebra wire-format invariants (SOAPAction contract segment, RequestHeader order, empty-Fields/populated-Filter) plus per-tool WSDL member order for every request builder — the failure mode for all of these is a silent empty result, not an error, so the tests are the only fast feedback loop.

Roadmap

• Live smoke-test suite for the 0.4.0 write shapes against a sandbox practice (shapes are WSDL-derived and unit-pinned; production verification is the remaining step)
• Client-side pagination (limit/offset + has_more) on the large list tools, mirroring tebra_get_all_patients
• Zod runtime validation with .strict() schemas so misspelled arguments fail loudly instead of being dropped
• Migration from Server.setRequestHandler to the SDK's McpServer/registerTool API, adding outputSchema/structuredContent
• Agent-facing evaluation set (10 read-only, verifiable questions) to catch wrong-but-plausible data — the failure class unit tests cannot see

Changelog

0.6.0 (2026-09-19)

Connectors Directory readiness. No wire-format changes.

• feat(annotations): every tool carries a title and MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) derived from its verb, pinned by tool-annotations.test.ts.
• feat(redact): PHI and credentials are scrubbed from the SOAP debug log, and patient identifiers from the request are scrubbed out of thrown error messages (SOAP faults, FHIR errors, and tebra_get_patient argument errors). Tested against synthetic records so no identifier survives.
• feat(fields): record-returning tools accept fields for minimum-necessary selection (dotted paths through arrays). tebra_get_all_patients defaults to a roster projection; tebra_get_patient omits insurance policy and group numbers unless named.
• refactor(fhir): handlers take a FhirConfig resolved once at startup, and the OAuth token cache is keyed per client.
• feat(mcpb): manifest.json (MCPB 0.2) with sensitive user_config for every credential, PRIVACY.md, npm run pack:mcpb, and a CI workflow that packs the bundle on version tags.
• fix(fhir): startup no longer exits when the FHIR key path is unreadable, and unexpanded Claude Desktop ${user_config.*} placeholders count as unset (verified live 2026-09-19: both took the desktop extension down as "Server disconnected" with no visible cause).
• docs: independence from Tebra and the BAA requirement stated in the README, package descriptions, and skill; server.json now marks TEBRA_SOAP_USER secret.

0.5.1 (2026-09-06)

• fix(retry safety): tebra_create_payment (and every other Create* action) is no longer re-sent after a timeout, a reset after send, or an HTTP 5xx. Those failures are ambiguous — Tebra may have committed the write with only the response lost — and Tebra's SOAP API has no idempotency key, so the automatic retry could post a second payment to a patient's account (#13). Create* actions now retry only on failures that provably happened before Tebra could act (429 throttle, connection refused/unresolved); anything else throws AmbiguousOutcomeError naming the read tool to verify with before resubmitting. Read actions keep the full 3-attempt retry.
• The "FHIR tools disabled" startup message and env-var doc comments now name the private_key_jwt route; package-lock.json resynced to the package version (#17).

0.5.0 (2026-09-04)

• feat(FHIR auth): SMART Backend Services private_key_jwt client authentication. With TEBRA_FHIR_PRIVATE_KEY_PATH + TEBRA_FHIR_KID set, the token request carries an RS384 client assertion (iss=sub=client id, aud=token URL, exp=5 min, unique jti per request) instead of a secret — the shape an appSphere Backend App registered with a JWKS URL requires. client_secret remains the fallback when no key path is set; an existing secret install is unchanged. FhirConfig.clientSecret is now optional and FhirConfig.privateKey is new. Tested offline: the assertion is verified against a published JWKS fixture, never a live token endpoint.

0.4.3 (2026-08-04)

• fix(external IDs): fail closed on IDs over 25 characters in tebra_create_patient and tebra_update_patient_external_id — Tebra's external-ID storage silently truncates at 25 chars (verified live via a UNIQUE KEY collision on the truncated value), which breaks every later lookup by the full value. IDs are also unique per vendor. (This guard just missed the 0.4.2 tarball.)

0.4.2 (2026-08-04)

The write path is now production-verified: iterating a live write-smoke harness against a real practice confirmed CreatePatient, CreateAppointment, CreateDocument, and the external-ID batch + vendor-scoped lookup end-to-end (encounter write ops remain opt-in to verify). The runs surfaced four additional wire facts, all fixed here:

• Required members (minOccurs audit after CreatePatient faulted "Expecting element 'Practice'"): PatientCreate/PatientUpdate require Practice; AppointmentCreate requires PracticeId; AppointmentUpdate requires PatientId+ServiceLocationId; DocumentCreateRequest requires PracticeId.
• External IDs are vendor-scoped, unique per vendor, and silently truncated at 25 characters — writes with longer IDs now fail closed (truncation breaks every later lookup).
• DocumentDate is a true xs:dateTime — date-only input is normalized to ISO midnight; US-format strings fault the deserializer (Filter date members, being xs:string, still accept either).
• The UpdatePatientsExternalID response returns an empty ItemsUpdated echo even on success — verify via the vendor-scoped ExternalID lookup on tebra_get_patient instead.

Details:

• fix(create/update_patient): Practice is a required member of PatientCreate AND PatientUpdate — now always emitted; when practiceName/practiceId are omitted, the account's first practice is auto-resolved via GetPractices (cached). tebra_update_patient gains optional practiceName/practiceId args.
• fix(create_appointment): PracticeId is required — auto-resolved when omitted.
• fix(update_appointment): PatientId and ServiceLocationId are required — auto-hydrated from GetAppointment when omitted, so a status-only or reschedule-only update still works.
• fix(create_document): PracticeId is required — auto-resolved when omitted.
• Builders fail closed with clear messages if the required members are still missing; 3 new regression tests (39 total).

0.4.1 (2026-08-04)

Live production verification of the 0.4.0 shapes — a full read-only smoke pass (24 checks: every SOAP read tool plus the FHIR pipeline) now passes against a real Tebra practice. Fixes found only by going live:

• fix(xsd7 namespace): GetServiceLocations and GetProcedureCodes faulted with "Expecting element 'Fields'" — xsd7's targetNamespace has NO trailing slash (…/api/schemas vs xsd0's …/api/schemas/), so their Fields/Filter members are different XML names. The envelope now declares both namespaces and those two tools emit kar7:-prefixed members.
• fix(get_appointment_reasons): PracticeId is required by the WSDL (fault when omitted). The handler now auto-resolves the account's first practice ID via GetPractices (cached) when not supplied.
• fix(search_patients): ToDateOfBirth is exclusive server-side — an exact dateOfBirth search now sends [DOB, DOB+1) instead of a zero-width range that matched nothing.
• fix(rate limiting): +250ms safety margin per endpoint window — an exact-interval gap still trips Tebra's server-side 429; throttle errors (reported inside HTTP-200 ErrorResponse blocks) are now classified retryable.
• fix(get_encounter): GetEncounterDetails returns EncounterStatus as a 1-based numeric code; now mapped to labels (3=Approved, verified live against the same encounter's charge rows).
• fix(FHIR auth): on invalid_scope, the token server names the scope the client is registered with — the client now retries once with that scope automatically (registrations vary between system/*.read and patient/*.read).

0.4.0 (2026-08-03)

Full-surface WSDL contract audit. Every request builder was re-derived from the live WSDL (KareoServices.svc?xsd=xsd0/xsd7), which surfaced that the Fields/Filter misplacement fixed for GetCharges in 0.3.0 affected every other list GET, and that most write operations used wrong wrapper elements or member names. The failure mode in all cases is silent (unfiltered results, dropped fields, or server-side faults), which is why these survived so long. 26 new regression tests pin the corrected shapes.

• fix(list GETs): patients, appointments, payments, transactions, providers, service-locations, procedure-codes, bulk-patients, and encounter-details now put criteria in <kar:Filter> (WSDL sequence order) with an empty <kar:Fields/> — previously all of their filter args were silently ignored, and every call returned the unfiltered set. Args with no WSDL filter member now fail closed with guidance (patients.mrn, appointments.providerId → use resourceName, payments.patientId).
• fix(get_patient / authorizations / eligibility): GetPatientReq has no Fields member at all — the ID now goes in Filter (SinglePatientFilter), so single-patient lookup works. tebra_get_patient also gains lookup by externalId/externalVendorId.
• fix(response parsers): real WSDL member names throughout — nested case/policy/authorization data (PatientCaseData → PatientInsurancePolicyData → PatientInsurancePolicyAuthorizationData, AuthorizedNumberOfVisits), MedicalRecordNumber, NationalProviderIdentifier, PatientBatchData + nextStartKey for bulk paging, ThrottleDetail, ExternalVendorData, EncounterDetailsData. Previous names matched nothing, so insurance/auth/case data always parsed empty. All list parsers drop Tebra's phantom empty placeholder row.
• fix(writes): create/update_patient members re-ordered to WSDL sequence (out-of-order members were silently dropped — DOB, gender, email, address never persisted) with correct names (SocialSecurityNumber, PatientExternalID, MedicalRecordNumber) and insurance nested under Cases → Policies; create/update_appointment rewritten to the flat AppointmentCreate/Update shape (StartTime/EndTime, PatientSummary, ProviderId); delete_appointment wraps Appointment; create_encounter rewritten to EncounterCreate (per-service-line DiagnosisCode1–4, RenderingProvider, Practice); update_encounter_status wraps EncounterUpdateStatus with the real enum (Draft/Submitted/Approved/Rejected/Unpayable — "Review" is a UI label, not an API status); create_payment uses the PaymentCreate nested groups; create_document wraps DocumentToCreate in member order; delete_document sends DocumentId (case-sensitive); external-ID tools use ExternalVendor / Updates→UpdateBatch batch shapes; create_appointment_reason in member order with integer color; validate_connection sends the credentials GetCustomerIdFromKeyRequest actually expects.
• change: tebra_update_patient_case → tebra_set_primary_patient_case — the underlying op (UpdatePrimaryPatientCase) only promotes a case to primary by PatientCaseId; it never accepted name/payer-scenario edits. Old tool name still routes.
• new: tebra_update_appointment_status — targeted status changes (CheckedIn/NoShow/Cancelled...) via UpdateAppointmentStatus, cheaper than a full update.
• fix(FHIR): default base URL corrected to https://fhir.prd.cloud.tebra.com/fhir-request (hyphen — the old /fhir/request path returns HTTP 200 with an empty body for every call, verified live 2026-08-03); empty-200 responses now raise a descriptive configuration error; 401 triggers one automatic token refresh + retry; from/to date ranges emit two repeated date params (previously percent-encoded into one malformed value); Bundle pagination followed automatically (link[rel=next], up to 10 pages, truncation flagged); required second search params defaulted (MedicationRequest intent=order, CarePlan category=assess-plan, CareTeam status=active) — without them Tebra returns a silent empty bundle.
• new: tebra_fhir_search_patients — FHIR Patient lookup by name/birthdate/identifier; FHIR patient IDs are a separate identifier space from SOAP IDs and previously had no in-server resolution path.
• fix(core): extractTag/extractAllTags require a tag-name boundary (Patient no longer matches PatientData); server version is read from package.json (was hardcoded 0.2.5); server name corrected to tebra-mcp-server; rate-limit table keys CreatePayment/UpdateAppointmentStatus.

0.3.2 (2026-07-15)

• chore: removed the bundled project-specific integration templates (src/integrations/epic-notes-integration.ts, src/integrations/fal-integration.ts) and every reference to them (README "Integration Services" section, workflow-example labels, contributor docs). These were copy-paste connector modules for external downstream projects — never imported by the server and not part of its runtime — so they did not belong in the published package. No MCP tools were added or removed; the 45-tool surface is unchanged and the compiled tarball no longer ships dist/integrations/.

0.3.1 (2026-07-07)

• fix(get_charges): <kar:Fields/> is now sent EMPTY. 0.3.0's explicit column toggles triggered Tebra's projection-inversion quirk — every call returned a single empty <ChargeData/> placeholder (zero real fields, no fault) regardless of filter matches. Empty Fields returns the full record, including the PrimaryInsurance* adjudication columns (payment, contract adjustment + reason, secondary adjustment + reason, adjudication date) the explicit toggles were meant to surface.
• Response parser (parseChargeBlocks, exported) drops the empty placeholder block — a no-match response previously counted as one phantom all-empty charge.
• status filter documented with live-observed enum values: Pending, Completed, Error - Rejection, Voided, Ready. At least some accounts have NO Denied status — denials surface as Error - Rejection.
• Verified live against production Tebra: a 12-month, 14-window pull returned 293 real charges with populated financials; server-side Status filtering confirmed working.

0.3.0 (2026-07-06)

• fix(get_charges): filter criteria moved out of <kar:Fields> into <kar:Filter>, emitted in WSDL (xsd0) sequence order — inside Fields they were silently skipped by WCF, so GetCharges returned unfiltered data for the package's entire history. patientId now throws (ChargeFilter has no patient ID member; use patientName).
• Note: Tebra enforces a server-side ≤60-day posting-date window on GetCharges.

0.2.6 (2026-07-06)

• fix(get_appointment_detail): GetAppointment does NOT take the Fields/Filter request shape the list endpoints use. Per the live WSDL (KareoServices.svc?xsd=xsd0), GetAppointmentReq = RequestBase + <Appointment>{ AppointmentId: xs:long } (lowercase "d"). The old envelope faulted on every call with 'EndElement' 'request' … Expecting element 'Appointment', so tebra_get_appointment_detail never worked against live Tebra.
• The response is the WSDL AppointmentCreate shape — patient nested under <PatientSummary> (group attendees under <PatientSummaries>), ISO StartTime/EndTime, AppointmentStatus, enum-letter AppointmentType. The tool's output now maps these real fields (patient name + DOB, appointment mode, reason id, recurrence, group attendees, audit timestamps) instead of the fictional AppointmentData field set.
• Verified live: full detail returned for a real appointment. Contributor lesson: single-record Tebra ops can have entirely different WSDL contracts from their list counterparts — read ?xsd=xsd0 before assuming the Fields/Filter pattern.

0.2.5 (2026-04-28)

• fix(soap): every GET request body now includes a sibling <kar:Filter /> after <kar:Fields>. Tebra's WSDL marks Filter as minOccurs="0", but their server-side GetFilteredX(...) methods dereference the filter parameter without null-checking and throw NullReferenceException when it's absent. Tools patched: practices, providers, service-locations, procedure-codes, transactions, payments, charges, encounters, patients (search + get-by-id), bulk-patients, appointments. Without 0.2.5, every GET call fails with a server-side NullRef.
• Added regression test asserting <kar:Filter /> is emitted in WSDL-required order (Fields before Filter).

0.2.4 (2026-04-28)

• fix(soap): <RequestHeader> children now serialize in WSDL-required order (CustomerKey → Password → User). The previous CustomerKey → User → Password order caused silent authorization failures even with valid credentials. Confirmed in writing by Tebra customer care.

0.2.3 (2026-04-28)

• fix(soap): SOAPAction HTTP header now includes the KareoServices/ WCF contract segment that Kareo's dispatcher requires. Versions 0.2.2 and earlier sent ${SOAP_NAMESPACE}${operation}, which the dispatcher rejected with HTTP 500 (ContractFilter mismatch at the EndpointDispatcher). Header value is now also explicitly quoted per RFC 3902 §3.2.
• Added a regression test asserting the exact header value (npm test).

0.2.2 (2026-04-27)

• fix(soap): RequestHeader (User/Password/CustomerKey) is now placed inside the request body where Tebra's WSDL expects it, rather than in the SOAP envelope header.

0.2.1 (2026-04-26)

• Published to the MCP Registry under com.jamesrosingmd/tebra (verified-domain namespace).

0.2.0

• Initial public release.

Upgrade urgently from 0.2.4 or earlier. All prior releases hit at least one of the three wire-format bugs above, and only 0.2.5 satisfies all three of Tebra's WSDL/runtime requirements end-to-end.

License

MIT License

Copyright (c) 2026 James H. Rosing, MD, FACS

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.