The AI price record, inside your agentTen tools, no key, the source behind every price
Free to call. No account, no API key, nothing to install.
> Add the Vitna MCP server to Claude Code. It is a remote Streamable HTTP server at https://vitna.ai/api/v1/mcp and needs no key, header or OAuth. Keep all ten tools. When it connects, run tools/list, then: Call vitna_get_market_tape with class "model" and list the five cheapest rows.Paste into Claude Code. It edits its own config, then checks the connection.
Read the web and 16 official sourcesfrom one hosted MCP endpoint
25 tools over Vitna's hosted Web APIs at web.vitna.ai/v1/mcp, with google_search routed through Vitna Search on an org key.
{ "url": "https://developers.openai.com/api/pricing", "title": "Pricing - OpenAI API", "format": "markdown", "content": "## Text tokens\n| Model | Input | Output |\n| gpt-5.4 | $2.50 | $10.00 | ...", "links": ["https://developers.openai.com/api/docs/models", "..."], "truncated": false, "injection": { "detected": false, "findings": [] }}
Response shape from the server's tool schemas. Full contracts in mcp.json.
claude mcp add --transport http vitna-web https://web.vitna.ai/v1/mcp --header "Authorization: Bearer $VITNA_WEB_KEY"16 official sources: Federal Register, eCFR, USAspending, SEC EDGAR, openFDA, ClinicalTrials, CISA KEV, FRED, FINRA, Treasury, NY Fed, BLS, CourtListener, FCC, EIA, DOL H-1B
One URL for every clientthat speaks MCP
Pick a client and the setup above rewrites itself.
vitna_check_quoteIs this quote fair?vitna_get_market_tapeWhat does it cost now?vitna_get_sku_historyWhat did it cost before?vitna_get_provider_termsWhat do they keep?vitna_cite_observationWhere is that from?vitna_get_sku_capabilitiesWhat can it do?vitna_decide_stackShould we switch?vitna_check_rolloutIs the rollout healthy?vitna_analyze_pipelineWhat does the pipeline cost?vitna_verify_audit_packDoes this pack verify?Free to call,traceable to the page
gpt-5.4's input price, and everything that comes with it.
/api/docs/pricing
22 Nov 14:08 UTC
the 22 Nov tree is not signed yet
LiteLLM 2.5e-06 a token
Built for agents that spend moneyIt checks the record first
Questions on the way to a purchase, and what the server answered to this page.
Measured against developers.openai.com as confirmed at 22 Nov 14:08 UTC. No peer spread: only 1 provider sells gpt-5.4.
Questions
What is Vitna MCP?
A hosted MCP server at https://vitna.ai/api/v1/mcp. It gives any agent ten read-only tools over Vitna's record of AI prices, in Claude Code, Claude, Cursor, Codex, VS Code or any client that connects to a remote Streamable HTTP server.
Is it free? Do I need an API key?
It is free, and there is nothing to sign up for: no account, key, header or OAuth step. Every tool answers anonymous calls.
Which tool should my agent call?
Most agents pick on their own from the descriptions. Roughly:
vitna_check_quotecompares a quoted rate with the list price and, where three or more providers sell the product, the peer spread.vitna_get_market_tapereads current list prices for one class, optionally for one provider.vitna_get_sku_historyreturns every price period a SKU has had, with the page and hash behind each.vitna_get_provider_termsreads retention, training, region and key-handling terms. The terms record holds no entries yet, so for now it answers with an empty list.vitna_cite_observationturns a figure's observation id into a reference, a footnote or a contract schedule line.vitna_get_sku_capabilitiesreturns three-provenance SKU capability and speed profiles.vitna_decide_stackweighs candidate models against an incumbent on cost per successful unit, using quality scores you send.vitna_check_rolloutevaluates standing-decision drift, shadow rule promotion readiness, and routed canary health.vitna_analyze_pipelinecomputes a multi-stage, multi-class AI pipeline Bill of Materials and spend optimization.vitna_verify_audit_packverifies avitna.audit-pack.v1procurement and governance attestation pack across six layers.
What does Vitna keep from my calls?
Quote checks and stack decisions are computed in memory, and the inputs are not stored. Every tool call is answered with Cache-Control: no-store, and every tool declares readOnlyHint: true, so no call changes anything on the server.
Where do the numbers come from?
From each provider's own pricing page, read on a schedule. Each read archives the page and records its SHA-256, and each day's reads are folded into a Merkle root signed with Ed25519. Verify a figure shows how to check one yourself.
No provider signs its own pricing page: the signature covers Vitna's record of the fetch and not the provider's page.
How fresh is a figure?
Every row carries the time its source last confirmed it. A row older than two of its source's read cycles is flagged stale, and the tools return the flag with the row rather than dropping it.
What happens when a call goes wrong?
An unknown tool name gets JSON-RPC error -32602. A bad argument, such as a class the record does not carry, comes back as a normal result with isError: true and a code like unknown_class, so the agent can read it and try again.
Can I call it from my own code?
Yes. It is plain JSON-RPC 2.0 over HTTPS POST, so curl works (pick Your own code in the setup panel for the exact call), and so does any MCP client library.
Where is Vitna's Web API and official data MCP server?
Web page reads, structural diffs, PDF/Office parsing, company dossiers, enrichment grids, claim verification, Google SERP capture, and 16 official regulatory data sources run on Vitna's sibling 25-tool MCP server at https://web.vitna.ai/v1/mcp (configured in Web & Sources MCP above) using a vitna_web_ or vitna_key_ bearer token. Try the live endpoints at /apis or inspect /.well-known/mcp.json.
Put the record in your agentFree prices. Hosted web & sources.
# Vitna MCP > First-party attested AI infrastructure pricing, compliance terms, cryptographic citations, and stateless stack decisions from Vitna Witness. Endpoint: https://vitna.ai/api/v1/mcp Transport: Streamable HTTP, JSON-RPC 2.0, MCP 2025-03-26 Auth: none Server: vitna-mcp 1.0.0 Discovery: GET https://vitna.ai/api/v1/mcp returns the descriptor and tool schemas as JSON ## Tools (10, all readOnlyHint: true) - vitna_check_quote(sku, quotedAmountUsd, dimension?, date?, promptTokens?): Evaluate a vendor AI pricing quote in memory against Witness list prices and peer spreads with zero retention. - vitna_get_market_tape(class?, provider?): Read current first-party attested list prices from the Witness market tape, optionally filtered to one component class and provider. - vitna_get_sku_history(slug, dimension?): Retrieve the append-only price period history for a Witness SKU, optionally narrowed to a single price dimension. - vitna_get_provider_terms(provider?, kind?): Inspect curated, source-verified provider data handling terms (zero_retention, no_training, regions, byok). - vitna_cite_observation(id): Resolve a Witness observation id into its archived document SHA-256, Ed25519 signed-day status, and copy-ready citations. - vitna_get_sku_capabilities(slug): Inspect a Witness SKU's three-provenance capability profile: mirrored catalog capabilities and context length, third-party speed and TTFT measurements, curated provider terms, and admissibility boundaries. - vitna_decide_stack(workload, candidates, evidence?): Run a stateless, zero-retention stack migration check comparing candidate SKUs against an incumbent SKU, compliance requirements, three-provenance capability and speed bounds, and inline eval scores. - vitna_check_rollout(workload, standingDecision?, rules?, ledgerRows?, observations?, evidence?, thresholds?): Evaluate standing-decision drift, shadow rule promotion readiness, and routed canary health in memory against Witness list prices, compliance terms, three-provenance capabilities, and inline runtime telemetry with zero retention. - vitna_analyze_pipeline(name?, runLabel?, monthlyRuns?, discoverPeers?, stages, privateRates?, ledgerRows?, evidence?): Compute a compound multi-stage AI pipeline Bill of Materials, per-run trace attribution, compound reliability, and cross-layer spend optimization across all 10 Witness component classes (model, compute, retrieval, speech, tooling, extraction, translation, image, video, music) with zero retention. - vitna_verify_audit_pack(mode?, pack?, stackfile?, trustedPackPublicKey?, trustedRecordPublicKey?, trustedPolicyPublicKey?, requirePackSignature?, requireAttestedPrices?, failOnRollback?, maxMonthlyCostUsd?, monthlyRuns?, discoverPeers?, policyText?, policyLock?, ledgerRows?, evidence?, terms?): Verify a cryptographic vitna.audit-pack.v1 procurement, FinOps, and governance attestation pack across all six layers, or assemble an inline audit pack from a Stackfile with zero retention. ## Setup prompt Add the Vitna MCP server to Claude Code. It is a remote Streamable HTTP server at https://vitna.ai/api/v1/mcp and needs no key, header or OAuth. Keep all ten tools. When it connects, run tools/list, then: Call vitna_get_market_tape with class "model" and list the five cheapest rows. ## Setup by client ### Claude Code # Register Vitna over Streamable HTTP. No key, no header. claude mcp add --transport http vitna https://vitna.ai/api/v1/mcp # Then ask in any session: # Call vitna_get_market_tape with class "model" and list the five cheapest rows. ### Claude # Claude on the web or desktop # Settings › Connectors › Add custom connector Name Vitna URL https://vitna.ai/api/v1/mcp # Leave the OAuth fields empty: the server is keyless. # Ask: # Call vitna_get_market_tape with class "model" and list the five cheapest rows. ### Cursor // .cursor/mcp.json, or ~/.cursor/mcp.json for every project { "mcpServers": { "vitna": { "url": "https://vitna.ai/api/v1/mcp" } } } ### Codex # ~/.codex/config.toml [mcp_servers.vitna] url = "https://vitna.ai/api/v1/mcp" ### VS Code // .vscode/mcp.json { "servers": { "vitna": { "type": "http", "url": "https://vitna.ai/api/v1/mcp" } } } ### LiteLLM # Price every request your proxy routes from Vitna's record. # LiteLLM loads the map from LITELLM_MODEL_COST_MAP_URL, # or from --model_cost_map_url with the same address. export LITELLM_MODEL_COST_MAP_URL=https://vitna.ai/api/v1/gateways/litellm/model-prices.json litellm --config config.yaml # The cost map is a price file rather than the MCP, # so the tool and class choices do not apply here. ### Your own code # List the tools curl -s https://vitna.ai/api/v1/mcp \ -H "content-type: application/json" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' # Call one curl -s https://vitna.ai/api/v1/mcp \ -H "content-type: application/json" \ -d '{"jsonrpc":"2.0","id":2,"method":"tools/call", "params":{"name":"vitna_get_market_tape","arguments":{"class":"model"}}}' ## Sibling server: Vitna Web, Search & Official Sources MCP Endpoint: https://web.vitna.ai/v1/mcp Transport: Streamable HTTP, JSON-RPC 2.0 Auth: Bearer token (vitna_web_... or vitna_key_... for google_search) Live workbench: https://vitna.ai/apis#try Unified MCP directory: https://vitna.ai/.well-known/mcp.json Web OpenAPI 3.1: https://web.vitna.ai/openapi.json ### Claude Code setup claude mcp add --transport http vitna-web https://web.vitna.ai/v1/mcp \ --header "Authorization: Bearer $VITNA_WEB_KEY" ### Tools (25) - find_sources: Find a data source - describe_source: Read a data source's contracts - run_source: Call a data source - read_page: Read a web page - batch_scrape: Read multiple web pages - extract_data: Extract data from pages - map_site: List a site's URLs - crawl_site: Crawl a site - get_crawl: Read a crawl's progress and pages - cancel_crawl: Stop a crawl - research_web: Research a question across web pages - list_open_roles: List a company's open roles - diff_pages: Compare pages or text snapshots - parse_document: Parse a PDF or Office document - verify_claim: Verify a factual claim - entity_dossier: Build a company dossier - enrich_grid: Enrich an entity matrix - investor_relations: Read an Investor Relations portal - list_firehose_events: Read structured public-registry events - capture_screenshot: Capture a browser screenshot - capture_pdf: Render a page to PDF - list_linked_documents: Find or download linked documents - search_index: Search Vitna's web index - discover_grid: Discover and verify structured entity grids - google_search: Search Google ### Official data sources (16) - Federal Register (federal-register): documents/search, documents/get - Electronic Code of Federal Regulations (ecfr): sections/search, sections/text - USAspending.gov (usaspending): awards/search, agencies/list - SEC EDGAR (sec-edgar): companies/lookup, filings/recent, facts/concept, filings/search - openFDA (openfda): drugs/labels, drugs/reactions, recalls/search, devices/events, drugs/recalls, devices/recalls - ClinicalTrials.gov (clinicaltrials): studies/search, studies/get - CISA Known Exploited Vulnerabilities (cisa-kev): catalog/search, catalog/summary - FRED (Federal Reserve Economic Data) (fred): series/observations - FINRA Reg SHO Short Sale Volume (finra): short-volume/ticker, short-volume/top - U.S. Treasury Fiscal Data (treasury): debt/daily, cash/tga, rates/average - Federal Reserve Bank of New York Markets (nyfed): rates/sofr, rates/effr, operations/rrp - Bureau of Labor Statistics (BLS) (bls): series/observations, series/latest - CourtListener (courtlistener): courts/list, opinions/search, dockets/search - FCC ECFS (Electronic Comment Filing System) (fcc): filings/search, proceedings/search - U.S. Energy Information Administration (EIA) (eia): energy/inventory, electricity/demand - U.S. H-1B & OFLC Talent Filings (dol-h1b): employers/approvals, lca/search ## Questions ### What is Vitna MCP? A hosted MCP server at https://vitna.ai/api/v1/mcp. It gives any agent ten read-only tools over Vitna's record of AI prices, in Claude Code, Claude, Cursor, Codex, VS Code or any client that connects to a remote Streamable HTTP server. ### Is it free? Do I need an API key? It is free, and there is nothing to sign up for: no account, key, header or OAuth step. Every tool answers anonymous calls. ### Which tool should my agent call? Most agents pick on their own from the descriptions. Roughly: - vitna_check_quote compares a quoted rate with the list price and, where three or more providers sell the product, the peer spread. - vitna_get_market_tape reads current list prices for one class, optionally for one provider. - vitna_get_sku_history returns every price period a SKU has had, with the page and hash behind each. - vitna_get_provider_terms reads retention, training, region and key-handling terms. The terms record holds no entries yet, so for now it answers with an empty list. - vitna_cite_observation turns a figure's observation id into a reference, a footnote or a contract schedule line. - vitna_get_sku_capabilities returns three-provenance SKU capability and speed profiles. - vitna_decide_stack weighs candidate models against an incumbent on cost per successful unit, using quality scores you send. - vitna_check_rollout evaluates standing-decision drift, shadow rule promotion readiness, and routed canary health. - vitna_analyze_pipeline computes a multi-stage, multi-class AI pipeline Bill of Materials and spend optimization. - vitna_verify_audit_pack verifies a vitna.audit-pack.v1 procurement and governance attestation pack across six layers. ### What does Vitna keep from my calls? Quote checks and stack decisions are computed in memory, and the inputs are not stored. Every tool call is answered with Cache-Control: no-store, and every tool declares readOnlyHint: true, so no call changes anything on the server. ### Where do the numbers come from? From each provider's own pricing page, read on a schedule. Each read archives the page and records its SHA-256, and each day's reads are folded into a Merkle root signed with Ed25519. Verify a figure (https://vitna.ai/docs/verify-a-figure) shows how to check one yourself. No provider signs its own pricing page: the signature covers Vitna's record of the fetch and not the provider's page. ### How fresh is a figure? Every row carries the time its source last confirmed it. A row older than two of its source's read cycles is flagged stale, and the tools return the flag with the row rather than dropping it. ### What happens when a call goes wrong? An unknown tool name gets JSON-RPC error -32602. A bad argument, such as a class the record does not carry, comes back as a normal result with isError: true and a code like unknown_class, so the agent can read it and try again. ### Can I call it from my own code? Yes. It is plain JSON-RPC 2.0 over HTTPS POST, so curl works (pick Your own code in the setup panel for the exact call), and so does any MCP client library. ### Where is Vitna's Web API and official data MCP server? Web page reads, structural diffs, PDF/Office parsing, company dossiers, enrichment grids, claim verification, Google SERP capture, and 16 official regulatory data sources run on Vitna's sibling 25-tool MCP server at https://web.vitna.ai/v1/mcp (configured in Web & Sources MCP (#mc-web) above) using a vitna_web_ or vitna_key_ bearer token. Try the live endpoints at /apis (https://vitna.ai/apis#try) or inspect /.well-known/mcp.json (https://vitna.ai/.well-known/mcp.json).
