generated: '2026-08-15'
method: probed
source: https://general.veevavault.dev/mcp + live tools/list against https://docs.veevavault.dev/mcp
status: published
note: >-
  Veeva ships TWO distinct first-party MCP servers, documented at general.veevavault.dev/mcp.
  They are different products with different reach, so they are recorded as separate servers
  rather than folded into one entry.
deployment:
  mode: remote
  endpoint: https://docs.veevavault.dev/mcp
  auth: none
  verified: probed
  note: >-
    The `deployment` block describes the PUBLIC, anonymously reachable server (Vault
    Documentation MCP). The second server, Vault MCP Server, is also remote but its endpoint
    is tenant-templated (https://{host}/api/ai/mcp, where {host} is the customer's own Vault
    DNS) and requires a Bearer API access token, so it has no single URL an agent can reach
    without a customer tenant. Its own deployment block is recorded under servers[].
servers:
  - name: Vault Documentation MCP
    slug: veeva-vault-documentation-mcp
    docs: https://general.veevavault.dev/mcp/vault-documentation-mcp-server/overview
    deployment:
      mode: remote
      endpoint: https://docs.veevavault.dev/mcp
      auth: none
      verified: probed
    transport: streamable-http
    protocol_version: '2024-11-05'
    server_version: 3.4.0
    tool_count: 1
    tools_file: mcp/veeva-docs-mcp-tools.json
    x-evidence:
      fetched: '2026-08-15'
      url: https://docs.veevavault.dev/mcp
      initialize_status: 200
      tools_list_status: 200
      content_type: text/event-stream
      anonymous: true
    capabilities:
      tools: true
      resources: true
      prompts: true
      logging: true
    tools:
      - name: search_documentation
        readonly: true
        description: >-
          Search Veeva Vault developer and help documentation and return content chunks with
          source URL, title, and excerpt.
        inputs: [query, source, release, app_family]
        sources:
          - vault_developer_documentation
          - vault_api_reference
          - vault_java_sdk_javadocs
          - vapil_javadocs
          - vault_help_documentation
        app_families: [platform, clinical, commercial, medical, quality, regulatory, qualityone, safety, vault_crm]
    install_examples:
      - client: Claude Code
        command: 'claude mcp add --transport http vault-documentation-mcp https://docs.veevavault.dev/mcp --scope user'
        source: https://general.veevavault.dev/mcp/vault-documentation-mcp-server/guides/claude-code
      - client: Gemini CLI
        source: https://general.veevavault.dev/mcp/vault-documentation-mcp-server/guides/gemini-cli
      - client: Gemini Code Assist
        source: https://general.veevavault.dev/mcp/vault-documentation-mcp-server/guides/gemini-code-assist
  - name: Vault MCP Server
    slug: veeva-vault-mcp
    docs: https://general.veevavault.dev/mcp/vault-mcp-server/overview
    deployment:
      mode: remote
      endpoint: 'https://{host}/api/ai/mcp'
      auth: api-key
      verified: searched
      note: >-
        The {host} placeholder is the customer's own Vault DNS, e.g. myvault.veevavault.com. There is no
        shared endpoint; each Vault tenant serves its own. Not probed — reaching it requires
        a customer Vault and an API access token, and no anonymous instance exists.
    transport: streamable-http
    protocol_version: '2024-11-05 and later'
    gated: true
    entitlement: Requires Vault AI enabled on the customer's Vault.
    authentication:
      scheme: Bearer
      credential: Vault API access token (value prefixed `veeva-vault-`)
      header: 'Authorization: Bearer {access_token}'
      oauth_supported: false
      basic_supported: false
      session_lifetime: 8 hours from initialization
      docs: https://general.veevavault.dev/mcp/vault-mcp-server/getting-started/authentication
    session:
      init: POST initialize -> server returns Mcp-Session-Id header
      then: POST notifications/initialized -> 202 Accepted
      header: Mcp-Session-Id required on all subsequent requests
      docs: https://general.veevavault.dev/mcp/vault-mcp-server/guides/initializing-a-session
    tool_model: >-
      Tools are NOT a fixed catalog. Each Vault AI agent action the authenticated user may
      execute is exposed as one MCP tool, named {agent_name}-{action_name}. The list is
      per-user and per-tenant; configuration changes take up to 8 hours to appear. Call
      tools/list after each session init. Custom agent actions configured by the customer
      appear as tools alongside the Veeva-supplied ones.
    supports:
      tools: true
      resources: false
      prompts: false
    documented_tools:
      - name: base_document_chat_agent__sys-summarize_document__sys
        title: Summarize Document
        required: [document_id]
        optional: [user_message]
      - name: base_document_chat_agent__sys-ask_questions_on_document__sys
        title: Ask Questions on Document
        required: [document_id]
        optional: [user_message]
      - note: >-
          Only these two are shown in the published tools/list example. The live list is
          tenant-specific and requires authenticated introspection; nothing further is
          recorded here rather than guessed.
    long_running:
      tcp_max: 5 minutes
      resume: GET with Mcp-Session-Id + Last-Event-ID, 30s hold, retry up to 1 hour
      cancel: notifications/cancelled with the original JSON-RPC requestId
    permissions:
      - 'Application: API: Access API'
      - 'Agents: Execute'
    errors_doc: https://general.veevavault.dev/mcp/vault-mcp-server/reference/errors
    install_examples:
      - client: Claude Code
        command: 'claude mcp add --transport http vault-mcp https://{host}/api/ai/mcp --header "Authorization: Bearer {access_token}"'
        source: https://general.veevavault.dev/mcp/vault-mcp-server/getting-started/connect-claude-code