generated: '2026-08-29' method: searched status: published source: >- https://docs.getmontecarlo.com/docs/mcp-server, https://docs.getmontecarlo.com/docs/mcp-server-reference-and-operations, https://docs.getmontecarlo.com/docs/monte-carlo-mcp-server-technical-overview provider: Monte Carlo providerId: monte-carlo-data name: Monte Carlo MCP Server description: >- A first-party, fully hosted Model Context Protocol server that exposes Monte Carlo's data + AI observability platform to agents — alerts, monitors, assets, lineage, query and job performance, and AI agent traces. Runs on AWS Lambda behind an integration gateway that authenticates every request and forwards it to Monte Carlo's GraphQL API, executing under the calling user's identity so existing RBAC applies. Stateless: no customer data is persisted or cached by the server. deployment: mode: remote endpoint: https://mcp.getmontecarlo.com/mcp auth: oauth verified: probed probe: method: POST tools/list url: https://mcp.getmontecarlo.com/mcp http_status: 401 www_authenticate: >- Bearer error="invalid_token", error_description="Authentication required", resource_metadata="https://mcp.getmontecarlo.com/.well-known/oauth-protected-resource/mcp", scope="offline_access" date: '2026-08-29' note: >- A conformant RFC 9728 challenge. The endpoint is live and reachable; the live tool schemas are auth-gated, so the tools[] below are the names Monte Carlo itself publishes in its docs and in its own open-source Agent Skills, NOT an introspected tools/list response. inputSchema for each tool requires an authenticated introspection we deliberately did not attempt. install_note: >- mode is `remote`, not `both`. Monte Carlo publishes no stdio MCP package of its own — the key-based path in the docs runs the THIRD-PARTY `npx -y mcp-remote https://mcp.getmontecarlo.com/mcp` bridge against the same hosted endpoint, so it is a client shim for the remote server rather than a Monte Carlo stdio server. transport: streamable-http protocol: MCP (HTTP transport, OAuth 2.1) authentication: - method: oauth2.1-dynamic-client-registration preferred: true note: >- Clients self-register via DCR (RFC 7591) at https://auth.getmontecarlo.com/oauth2/register — there is no static client ID or secret to create. Scope offline_access. Requires an Editor role or above in Monte Carlo. authorization_server: https://auth.getmontecarlo.com protected_resource_metadata: https://mcp.getmontecarlo.com/.well-known/oauth-protected-resource/mcp - method: verified-connector note: >- Monte Carlo ships as an Anthropic-verified connector in the Claude Connectors Directory (https://claude.com/connectors/monte-carlo) — no keys to manage. - method: mcp-server-key note: >- Scoped MCP-server-only credentials (key id + secret) for service accounts and CI, sent as `Authorization: Bearer :` or as x-mcd-id / x-mcd-token headers. Standard Monte Carlo API keys do NOT work against the MCP server. Created via UI, `montecarlo mcp create-key`, or the createMcpIntegrationKey GraphQL mutation. clients_documented: - Claude / Claude Desktop (verified connector) - Claude Code (claude mcp add --transport http monte-carlo-mcp https://mcp.getmontecarlo.com/mcp) - Cursor - VS Code - Snowflake Cortex Agents (CREATE EXTERNAL MCP SERVER, OAUTH_DYNAMIC_CLIENT) toolsets: - name: default always_on: true description: Alerts, monitors, assets, lineage, data operations, utility tools. - name: extended header: 'x-mcd-toolset: extended' description: Adds use-case and dashboard-widget tools. - name: agent_observability header: 'x-mcd-toolset: agent_observability' description: Adds AI agent inspection tools (metadata, conversations, traces). prompts: published: true capability: MCP prompts note: >- The server publishes curated workflow prompts over the MCP prompts capability; supported clients discover them on connect and surface them as /mcp____. items: - name: mc-asset-health description: Check the current health of a table before querying it. - name: mc-triage description: Triage Monte Carlo alerts so you know where to focus. - name: mc-incident-response description: Full incident orchestration — triage through root cause to remediation. - name: mc-proactive-monitoring description: Discover coverage gaps and create monitors to close them. tools_source: >- Tool names are taken verbatim from Monte Carlo's own documentation (docs.getmontecarlo.com/docs/mcp-server) and from the 19 Apache-2.0 Agent Skills the company publishes at https://github.com/monte-carlo-data/mc-agent-toolkit, which call these tools by name. Input schemas are NOT recorded because the live tools/list is auth-gated (401). tool_count: 56 tools: - name: create_llm_span toolset: agent_observability kind: write - name: create_or_update_agent_evaluation_monitor toolset: agent_observability kind: write - name: create_or_update_agent_metric_monitor toolset: agent_observability kind: write - name: create_or_update_agent_trajectory_monitor toolset: agent_observability kind: write - name: create_or_update_agent_validation_monitor toolset: agent_observability kind: write - name: create_or_update_alert_comment toolset: default kind: write - name: create_or_update_comparison_monitor toolset: default kind: write - name: create_or_update_data_product toolset: default kind: write - name: create_or_update_metric_monitor toolset: default kind: write - name: create_or_update_sql_monitor toolset: default kind: write - name: create_or_update_table_monitor toolset: default kind: write - name: create_or_update_table_monitor_asset_rule toolset: default kind: write - name: create_or_update_validation_monitor toolset: default kind: write - name: get_agent_conversation toolset: agent_observability kind: read - name: get_agent_conversations toolset: agent_observability kind: read - name: get_agent_metadata toolset: agent_observability kind: read - name: get_agent_segments toolset: agent_observability kind: read - name: get_agent_trace toolset: agent_observability kind: read - name: get_agent_traces toolset: agent_observability kind: read - name: get_alert_agent_classification toolset: default kind: read - name: get_alerts toolset: default kind: read - name: get_asset_lineage toolset: default kind: read - name: get_audiences toolset: default kind: read - name: get_change_timeline toolset: default kind: read - name: get_domains toolset: default kind: read - name: get_etl_issues toolset: default kind: read - name: get_etl_jobs toolset: default kind: read - name: get_field_lineage toolset: default kind: read - name: get_field_metric_definitions toolset: default kind: read - name: get_github_prs toolset: default kind: read - name: get_jobs_performance toolset: default kind: read - name: get_mc_webapp_url toolset: default kind: read - name: get_monitor_report toolset: default kind: read - name: get_monitors toolset: default kind: read - name: get_queries_for_table toolset: default kind: read - name: get_query_changes toolset: default kind: read - name: get_query_latency_distribution toolset: default kind: read - name: get_query_rca toolset: default kind: read - name: get_reinforcement_loop_report toolset: default kind: read - name: get_reinforcement_loop_summaries toolset: default kind: read - name: get_table toolset: default kind: read - name: get_table_freshness toolset: default kind: read - name: get_table_size_history toolset: default kind: read - name: get_tables_for_job toolset: default kind: read - name: get_tasks_performance toolset: default kind: read - name: get_top_slow_queries toolset: default kind: read - name: get_troubleshooting_agent_results toolset: default kind: read - name: get_unmonitored_tables_with_anomalies toolset: default kind: read - name: get_use_case_table_summary toolset: extended kind: read - name: get_use_case_tables toolset: extended kind: read - name: get_use_cases toolset: extended kind: read - name: get_validation_predicates toolset: default kind: read - name: get_warehouses toolset: default kind: read - name: run_troubleshooting_agent toolset: default kind: write - name: search toolset: default kind: read - name: update_alert toolset: default kind: write network_access_control: supported: true docs: https://docs.getmontecarlo.com/docs/network-access-control note: >- IP/CIDR allowlists can be scoped to the MCP surface specifically (scopes cover MCP, API, Ingestion, UI, Agent and OpenTelemetry). security: token_passthrough: false note: >- Monte Carlo states the MCP server never reuses customer tokens directly, avoiding the MCP token-passthrough anti-pattern; calls execute under the authenticated user's identity in the GraphQL API. TLS 1.2+ in transit; stateless, no caching of customer data. docs: https://docs.getmontecarlo.com/docs/ai-security