Backwork MCP¶

Official MCP server for Backwork, a Medicare and payer policy intelligence platform - research coverage, prior authorization, claim risk and drug formulary evidence from the assistant, with every fact carrying numbered citations back to the policy it came from. Eight workflow-level tools (deliberately not a 1:1 API wrapper), a read-only OAuth grant by default, MCP Apps cards in ChatGPT, and a Claude Code plugin that bundles four clinical-research skills. The healthcare-ops find of the October 10 evening sweep, catalogued with the hosted endpoint probed live (initialize returns 401, missing_bearer_token).

Server type: Remote (Streamable HTTP at https://backworkhealth.com/mcp) or local stdio via npx -y @backwork/mcp
Auth: OAuth in the browser (read-only grant unless scopes include write) or a Backwork API key (bwk_live_...)
Tools: 8 workflow tools (backwork_coverage_lookup, backwork_policy_research, backwork_claim_validation, backwork_prior_auth_research, backwork_drug_formulary_research, backwork_compliance_review, backwork_webhook_management, backwork_system_health)
Category: Healthcare Operations
Built by: Backwork (backworkhealth.com; repo github.com/tylergibbs1/backwork-mcp, MIT; registry io.github.tylergibbs1/backwork-mcp)

Why This Matters for Operators¶

Ask a clinic's staff where the rule for a denied claim lives and you get a shrug: it is split across Medicare LCDs and NCDs, each MAC's articles, and commercial payer policies, updated on their own schedules. Backwork turns that pile into a research surface - which policy lists these codes, what documentation a prior auth needs, what the denial risk is, whether a commercial formulary covers a drug - and grounds every answer in a named policy with citations. For practices, billing teams and consultants, it compresses the lookups that eat hours per claim into one conversation, and the compliance-review tool watches policy changes so nothing quietly moves underneath a workflow.

Tools & Capabilities¶

Tool	What it does
`backwork_coverage_lookup`	Look up procedure codes and combine code details, policy evidence, prior authorization, claim risk, jurisdiction comparison and spending evidence; with a `payer` filter, code details list only that payer's policies
`backwork_policy_research`	Actions: `compare` codes side by side across Medicare contractors (MACs) with coverage counts, `search` matching policies with payer and number, `get` one policy's summary and criteria, `criteria` matching excerpts, `changes` recent policy changes, `jurisdictions` each MAC and its states
`backwork_claim_validation`	Validate claim coverage, documentation requirements, denial risk and optional policy-specific criteria
`backwork_prior_auth_research`	Prior-auth determination and confidence, codes requiring authorization, documentation to gather, known gaps and numbered citations; can also run a background job that searches public payer websites
`backwork_drug_formulary_research`	Commercial pharmacy-benefit evidence from CVS Caremark, Express Scripts and UnitedHealthcare / Optum Rx
`backwork_compliance_review`	Compliance stats and unreviewed policy changes; with an API key, acknowledge changes
`backwork_webhook_management`	Manage webhook endpoints (one event, `compliance.acknowledged`); needs a write-scoped API key
`backwork_system_health`	API health and dependency status; API-key connections only

Installation¶

{
  "mcpServers": {
    "backwork": {
      "url": "https://backworkhealth.com/mcp"
    }
  }
}

Claude Code: claude mcp add --transport http backwork https://backworkhealth.com/mcp, then finish the browser login and consent. Claude Desktop, Cursor, VS Code and Codex configurations are documented, and there is a Claude Code plugin with four skills (prior authorization, coverage checks, policy changes, denial appeal prep): /plugin marketplace add tylergibbs1/backwork-mcp and /plugin install backwork@backwork. New accounts start with 100 free credits.

Configuration and Safety¶

OAuth grants are read-only unless scopes include write: write-scope actions (webhook management, compliance acknowledgements) are withheld from a read-only connection, and the hosted endpoint rejects a raw API key passed as a bearer token.
The local npx -y @backwork/mcp process uses BACKWORK_API_KEY (bwk_); the server validates or introspects every token per request and does not store OAuth tokens.
Results carry provenance: source_urls, authority, retrieved_at and as_of per fact, plus applicability notes; Medicare content carries a non-Medicare disclaimer when relevant and commercial policy headers where payer type matters.
Rate limits follow the API plan; the docs point at a higher-capacity plan when the reset window is hit.

Business Relevance¶

Medical practices and clinics: check coverage and prior-auth requirements before the visit or the claim, so staff know the documentation to gather while there is still time to gather it.
Billing companies and RCM teams: validate claims against the policy evidence and see denial risk with citations - and the policy-change feed means the rulebook moving no longer shows up first as a rejected claim.
Consultants and compliance leads: review policy changes with an acknowledgment trail, and pull formulary evidence from the three largest commercial PBMs in the same surface.

Integration with CorpusIQ¶

Policy answers and business answers complete each other. Ask Backwork what Medicare or a payer requires for a code, then ask CorpusIQ how your own numbers look - claim, revenue and schedule patterns from the systems you already connect - and keep both in one thread, each traceable to its source.

Limitations¶

The MCP server is a research and validation surface; it does not submit claims or prior authorizations, and the practice management system remains the system of record.
Prior-auth jobs against public payer websites run as background tasks and show as pending until polled again.
Some tools and actions are gated by credential scope (read-only OAuth hides webhook management and compliance acknowledgement; backwork_system_health is API-key only).
Coverage depth follows the platform's policy library; confirm anything high-stakes against the linked policy document, as with any research tool.

FAQ¶

Do I need an API key or is OAuth enough?¶

OAuth is enough for the research core - coverage, policy, claim, prior-auth and formulary tools - and grants read-only by default. An API key unlocks the extra actions: compliance acknowledgements and webhook management.

Does it write anything into our systems?¶

No. The tools research, validate and report; the only writes are inside Backwork itself (acknowledging a reviewed policy change, managing Backwork webhooks), and those need a write-scoped credential.

What sources does it cite?¶

Medicare LCDs, Articles and NCDs, MAC jurisdiction data, and commercial payer policy documents, with per-fact source_urls and retrieval timestamps; drug formulary evidence comes from CVS Caremark, Express Scripts and UnitedHealthcare / Optum Rx.