generated: '2026-09-19'
method: probed
status: published
source: https://witness.getvda.ai/api/witness/mcp
docs: https://witness.getvda.ai/docs
summary: 'Verified Digital Agents runs FOUR live remote MCP surfaces on its own hosts plus a 98-server metered fleet,
  all Streamable HTTP (POST JSON-RPC 2.0) and all probed anonymously on 2026-09-19. (1) Witness at https://witness.getvda.ai/api/witness/mcp
  — protocol 2024-11-05, serverInfo {vda-witness 1.0.0, vendor "Verified Digital Agents (VDA)"}, 16 tools with real
  inputSchemas returned to an anonymous tools/list; discovery, `verify`, `verify_record_issuer`, `check_valid` and
  free key minting (`get_test_key`) need no credential, sealing needs the Witness bearer key the server itself mints.
  (2) C2MD at https://c2md.getvda.ai/mcp — protocol 2024-11-05, serverInfo {c2md-compliance-agent 1.0.0}, 11 tools;
  initialize/tools/list/`assess_agent_risk`/`get_test_key` are anonymous, the generative tools need a Witness bearer
  whose account tier maps to the c2md:* scopes. (3) HITL at https://hitl.getvda.ai/mcp — GATED: initialize itself
  answers 401 {"error":"unauthorized","message":"missing bearer credential"}; its 9 tools (raise/list/get/resolve
  items, baselines, authority config, quota) are documented with inputSchemas in the served agent card and mirror
  the REST /v1/tools/* operations 1:1 in the OpenAPI. (4) The GOSCE Portfolio Router at https://router.getvda.ai/mcp
  — protocol 2026-07-28, serverInfo {router 0.1.0}, 2 tools (invoke, selftest); `invoke` answered anonymously with
  the capability index (the router''s free tier). Behind it sit 98 templated fleet servers (<agent>.getvda.ai/mcp,
  2 tools each, invoke metered at $0.025 USDC per call via x402 v2 or a Stripe card rail, selftest free), enumerated
  in well-known/getvda-ai-agents-ai-catalog.json. ACP declares an MCP surface as PLANNED in its agent card (mcp.status:
  planned) and POST /mcp on acp.getvda.ai is a 404 — it is recorded here as absent, not derived. Authorship is unambiguous:
  every serverInfo/vendor string, tool name and description is VDA-specific (seal_hitl_decision, issue_admission_credential,
  generate_dpia_fria_scaffold) and matches no shared-platform fingerprint.'
deployment:
  mode: remote
  endpoint: https://witness.getvda.ai/api/witness/mcp
  auth: none
  verified: probed
  note: 'Hosted HTTPS endpoints an MCP client POSTs to directly; VDA ships no stdio MCP package and documents no
    npx/uvx install (the npm/PyPI packages are SDKs and CI tooling, not MCP servers — see packages/). `auth: none`
    is the connection-level truth for Witness, C2MD and the router: initialize and tools/list answer with no credential
    and no OAuth metadata is served on any host (/.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource
    404 on witness, c2md, hitl and router). Invocation is then gated per tool: Witness/C2MD write tools take `Authorization:
    Bearer wtn.<keyId>.<secret>` (a key the Witness server mints in-band with no human), HITL requires that bearer
    for every method including initialize, and the fleet meters `invoke` with an HTTP 402 x402 challenge. Streaming
    transport was not exercised. GET behaviour of the endpoints (probed 2026-09-19): Witness GET /api/witness/mcp
    -> 404 (Express, POST-only); C2MD GET /mcp -> 200 application/json (107 bytes); router GET /mcp -> 405; HITL
    GET /mcp -> 401. A GET that is not 200 is not a dead pointer here — every one answers a POST.'
servers:
- id: witness
  name: VDA Witness
  endpoint: https://witness.getvda.ai/api/witness/mcp
  transport: streamable-http
  http_methods:
  - POST
  auth: none for discovery/verify/get_test_key; Witness bearer key (wtn.<keyId>.<secret>) for seal/read/report/whoami/revoke
  status: live
  openapi: openapi/getvda-ai-witness-openapi.json
  agent_card: a2a/getvda-ai-agent-card.json
  probe:
    fetched: '2026-09-19'
    initialize:
      http_status: 200
      protocol_version: '2024-11-05'
      server_info:
        name: vda-witness
        version: 1.0.0
        vendor: Verified Digital Agents (VDA)
      capabilities:
        tools: {}
      file: mcp/getvda-ai-witness-mcp-initialize.json
    tools_list:
      http_status: 200
      content_type: application/json; charset=utf-8
      tool_count: 16
      file: mcp/getvda-ai-witness-mcp-tools.json
  tools:
  - name: get_test_key
    auth: none
    description: Self-issue a SEALED-tier Witness API key in-band (free; 5
    000 seals/month; optionally bind an Ed25519 controller JWK for a durable account): null
  - name: renew_challenge
    auth: none
    description: Agent-provable renewal step 1 — one-time nonce for an accountId
  - name: renew_key
    auth: none (controller-key signature)
    description: Renewal step 2 — prove control of the account
    receive a fresh short-TTL key: null
  - name: seal
    auth: bearer
    description: General-purpose seal of a governed decision (compatibility; prefer a shaped seal)
  - name: seal_hitl_decision
    auth: bearer
    description: Seal a human-in-the-loop decision with evidentiary basis
  - name: seal_agent_action
    auth: bearer
    description: Seal an autonomous agent action under a governing rule
  - name: seal_attestation
    auth: bearer
    description: Seal a fact/state as of a time
  - name: issue_admission_credential
    auth: bearer
    description: Issue an agent admission credential (a sealed attestation)
  - name: check_valid
    auth: none
    description: Is an admission credential currently valid (PUBLIC)
  - name: revoke_admission_credential
    auth: bearer (issuing account only)
    description: Revoke an admission credential
  - name: verify
    auth: none
    description: Independently verify a record or chain (Ed25519 signature + hash chain)
  - name: report
    auth: bearer
    description: EU AI Act Article 12 evidence report from the sealed trail
  - name: list_records
    auth: bearer
    description: Paginated summaries of the account's own records
  - name: get_record
    auth: bearer
    description: Full signed body of one record + verification + anchor state
  - name: whoami
    auth: bearer
    description: Resolve the presented key to account
    tier: null
    scopes: null
    expiry (Contract A cross-service auth): null
  - name: verify_record_issuer
    auth: none
    description: Issuer-authenticity verdict for a record id (never the body)
- id: c2md
  name: C2MD Compliance Agent
  endpoint: https://c2md.getvda.ai/mcp
  transport: streamable-http
  http_methods:
  - POST
  auth: none for initialize/tools/list/assess_agent_risk/get_test_key; Witness bearer key for generative tools (account
    tier maps to c2md:* scopes)
  status: live
  agent_card: a2a/getvda-ai-c2md-agent-card.json
  a2a_endpoint: https://c2md.getvda.ai/a2a
  probe:
    fetched: '2026-09-19'
    initialize:
      http_status: 200
      protocol_version: '2024-11-05'
      server_info:
        name: c2md-compliance-agent
        version: 1.0.0
      capabilities:
        tools:
          listChanged: false
      instructions: C2MD generates EU AI Act / GDPR / NIST / ISO 42001 governance artifacts for AI agents. Discovery
        and `assess_agent_risk` are free and need no key. Everything else needs a getvda.ai suite key — call `get_test_key`
        to find out how to get one free. Generated artifacts are DRAFTS requiring compliance-officer sign-off; C2MD
        does not deliver legal compliance.
      file: mcp/getvda-ai-c2md-mcp-initialize.json
    tools_list:
      http_status: 200
      content_type: application/json
      tool_count: 11
      file: mcp/getvda-ai-c2md-mcp-tools.json
  tools:
  - name: assess_agent_risk
    auth: none
    description: Cross-framework (EU AI Act + GDPR) risk assessment of an agent description; diagnostic only
  - name: extract_governance_inputs
    auth: none
    description: Derive reviewable governance inputs from an agent's own A2A card (genesis-from-card)
  - name: generate_admission_attestation
    auth: bearer
    description: Compliance attestation for an admission credential from its Witness chain
  - name: generate_compliance_bundle
    auth: bearer (c2md:generate_starter / generate_pro)
    description: AGENTS.md + SOP.md + SKILL.md governance bundle; long-running (task descriptor)
  - name: generate_dpia_fria_scaffold
    auth: bearer (c2md:generate_pro)
    description: GDPR Art. 35 DPIA / EU AI Act Art. 27 FRIA scaffold
  - name: generate_evidence_readiness_report
    auth: none in demo mode; bearer or an offline Witness proof bundle otherwise
    description: EU AI Act article-by-article evidence and readiness report
  - name: generate_journey_baseline
    auth: bearer (c2md:generate_journey)
    description: PLANNED — currently returns JSON-RPC -32601 method not found
  - name: list_supported_frameworks
    auth: none
    description: Supported frameworks
    jurisdictions and version anchors: null
  - name: translate_control
    auth: none
    description: One NIST/ISO/EU AI Act/GDPR control reference into MUST/MUST NOT/MAY rules
  - name: get_task
    auth: bearer
    description: Poll a long-running task by id
  - name: get_test_key
    auth: none
    description: Explains how to obtain a free suite key (minted by Witness
    not C2MD): null
- id: hitl
  name: HITL — human decisions on agent actions
  endpoint: https://hitl.getvda.ai/mcp
  transport: streamable-http
  http_methods:
  - POST
  auth: Witness bearer key (Contract A) required for EVERY method, initialize included
  status: live (gated)
  openapi: openapi/getvda-ai-hitl-openapi.json
  agent_card: a2a/getvda-ai-hitl-agent-card.json
  probe:
    fetched: '2026-09-19'
    initialize:
      http_status: 401
      content_type: application/json; charset=utf-8
      body: '{"error":"unauthorized","message":"missing bearer credential"}'
    tools_list:
      http_status: 401
      body: '{"error":"unauthorized","message":"missing bearer credential"}'
    note: The live schema is auth-gated. The 9 tool names and inputSchemas below are taken from the served, signed
      agent card (skills[].inputSchema) and the OpenAPI, which document POST /mcp as "initialize | tools/list |
      tools/call" and expose each tool as POST /v1/tools/<name>; they are not a live tools/list capture.
  tools:
  - name: raise_hitl_item
    rest: raise_hitl_item
    description: Raise a decision for human review
  - name: list_hitl_items
    rest: list_hitl_items
    description: List decision items (summaries carry no evidence)
  - name: get_hitl_item
    rest: get_hitl_item
    description: Get one decision item
  - name: resolve_hitl_item
    rest: resolve_hitl_item
    description: Record a human decision (sealed customer-managed to Witness)
  - name: list_baselines
    rest: list_baselines
    description: List active baselines for a caller
  - name: revoke_baseline
    rest: revoke_baseline
    description: Revoke a baseline (never a delete; sealed)
  - name: match_baseline
    rest: match_baseline
    description: Parity check for baseline containment
  - name: register_authority_config
    rest: register_authority_config
    description: Register an activated authority config
  - name: get_raise_quota
    rest: get_raise_quota
    description: Remaining raises this hour
    open-item headroom: null
    permitted decision classes: null
- id: gosce-router
  name: GOSCE Portfolio Router
  endpoint: https://router.getvda.ai/mcp
  transport: streamable-http
  http_methods:
  - POST
  auth: none; fleet `invoke` calls are metered with x402 v2 (HTTP 402 PAYMENT-REQUIRED, USDC on Base eip155:8453,
    or Stripe card via nvm:card-delegation)
  status: live
  openapi: openapi/getvda-ai-gosce-router-openapi.json
  agent_card: a2a/getvda-ai-gosce-router-agent-card.json
  probe:
    fetched: '2026-09-19'
    initialize:
      http_status: 200
      protocol_version: '2026-07-28'
      server_info:
        name: router
        version: 0.1.0
      capabilities:
        tools:
          listChanged: false
      file: mcp/getvda-ai-gosce-router-mcp-initialize.json
    tools_list:
      http_status: 200
      content_type: application/json
      tool_count: 2
      file: mcp/getvda-ai-gosce-router-mcp-tools.json
    tools_call_invoke:
      http_status: 200
      note: An anonymous tools/call of `invoke` with input "{}" returned the capability→agents index rather than
        a 402; the router treats its own routing answer as free.
  tools:
  - name: invoke
    description: Route an orchestrator to a verified fleet agent by capability
    with example output from the router's own selftest probes: null
  - name: selftest
    description: FREE — run the agent's real capability on canned input and return a graded result plus a trust
      manifest
  fleet:
    count: 98
    endpoint_pattern: https://<agent>.getvda.ai/mcp
    catalog: well-known/getvda-ai-agents-ai-catalog.json
    index: https://agents.getvda.ai/agents
    per_server_tools:
    - invoke
    - selftest
    free:
    - tools/list
    - server/discover
    - selftest
    price: $0.025 USDC per metered tools/call (health.getvda.ai health_check $0.15); not charged on failure
    sample_probe:
      endpoint: https://authenticated-llm-agent-cryptography.getvda.ai/mcp
      tools_list_http_status: 200
      tool_count: 2
      server_info:
        name: authenticated-llm-agent-cryptography
        version: 0.1.1
- id: acp
  name: VDA ACP
  endpoint: null
  status: planned
  probe:
    fetched: '2026-09-19'
    url: https://acp.getvda.ai/mcp
    http_status: 404
    body: '{"message":"Route POST:/mcp not found","error":"Not Found","statusCode":404}'
  note: The ACP agent card declares mcp.status "planned" ("An MCP surface (mirroring these skills as tools) will
    sit alongside this REST surface ... not yet live"). Nothing derived.