Referencing Reference
The MCP endpoint that grounds and verifies every claim your AI Engine produces.
Overview
Referencing is an MCP endpoint that any MCP-compatible LLM (Claude, ChatGPT, Gemini, Copilot) can connect to. It exposes a set of tools for verifying claims, extracting structured data, and managing a workspace-scoped library of trusted sources — all with a full audit trail.
Bearer tokens & scopes
All requests require an Authorization: Bearer <token> header. Tokens are minted per-workspace from the app and carry one or more veriref:* scopes:
veriref:verify— verify_* toolsveriref:extract— extract_* toolsveriref:library.read/veriref:library.write— library_* toolsveriref:admin— admin_* tools
Tools reference
| Tool | Description | Inputs |
|---|---|---|
| verify_run | Verify every claim in a passage against approved sources. | text, vertical?, source_ids? |
| verify_claim | Verify a single claim and return supporting/refuting sources. | claim, vertical? |
| verify_status | Poll the status of an async verify run. | run_id |
| verify_cancel | Cancel an in-flight verify run. | run_id |
| verify_report | Return the full audit report for a completed run. | run_id, format? |
| verify_sources | List sources considered for a run, with match scores. | run_id |
| extract_run | Extract structured claims, entities and citations from unstructured text. | text, schema? |
| extract_claims | Extract atomic factual claims for downstream verification. | text |
| extract_entities | Named-entity extraction (drug, company, ticker, statute…). | text, vertical? |
| extract_citations | Extract inline and footnote citations from a document. | text |
| extract_tables | Parse embedded tables into structured rows. | text|pdf_url |
| library_save | Save a verified source or clip to the workspace library. | source, tags?, notes? |
| library_search | Semantic + keyword search across saved library items. | query, filters? |
| library_get | Fetch a single library item with its full metadata. | item_id |
| library_update | Update tags, notes or verification status on an item. | item_id, patch |
| library_delete | Delete a library item (soft-deleted for 30 days). | item_id |
| library_list | List library items with pagination and filters. | cursor?, filters? |
| admin_usage | Return credit consumption and tool call counts. | period? |
| admin_scopes | List the caller's active scopes. | — |
| admin_keys | Manage bearer tokens for the workspace. | action, key_id? |
| admin_audit | Return the workspace audit trail. | since?, cursor? |
| admin_workspace | Read workspace settings (members, plan, region). | — |
Sources by vertical
Life Sciences
- PubMed
- Europe PMC
- openFDA
- ClinicalTrials.gov
- WHO ICTRP
Finance
- SEC EDGAR
- FRED
- IMF
- World Bank
- FDIC
- FCA
- Companies House
- FINRA BrokerCheck
Legal
- CourtListener
- Legislation.gov.uk
- EUR-Lex
- USPTO
- BAILII
Insurance
- NAIC
- EIOPA
- NHTSA
- FEMA
Rate limits & credits
Each plan includes a monthly credit allowance. One verify_run consumes credits based on the number of claims verified and sources retrieved. Rate limits are 60 requests/minute per token by default, raised on request for Business and Enterprise workspaces.
When you exceed a limit the endpoint responds 429 rate_limited with a Retry-After header. See Pricing for plan credit allowances.
Error codes
| 401 unauthorized | Missing or invalid bearer token. |
| 403 scope_denied | Token lacks the required veriref:* scope. |
| 404 not_found | The referenced run_id or item_id does not exist. |
| 409 conflict | Run is already terminal (completed/cancelled). |
| 422 invalid_input | Payload failed schema validation. |
| 429 rate_limited | Rate or credit limit exceeded — see Retry-After header. |
| 500 internal_error | Unexpected server error. Safe to retry with backoff. |
| 503 unavailable | A downstream source is temporarily unreachable. |
Changelog
2026-07-19 — Initial public reference published.