# MCP Tool Resolution
# Requires `--features openai-mcp-tools,store-sqlite` because these filters are opt-in.
#
# Demonstrates the `openai_mcp_tool_resolve` filter, which resolves MCP
# tool entries in the Responses API `tools` array into concrete
# tool definitions by calling `tools/list` on each upstream MCP
# server.
#
# The filter runs after `openai_tool_parse` (gated on `openai_tool_parse.has_mcp`)
# and before `openai_responses_proxy`. It reads MCP entries from the
# buffered request body, checks `previous_tools` for cached
# listings, calls `tools/list` via the `mcp_client` module, and
# writes `mcp_tool_map` to `ResponsesState`.
# For `stream: true`, runtime and response-processing failures are
# returned as a Responses SSE lifecycle containing
# `response.mcp_list_tools.failed` and terminal `response.failed` events.
# Local request-policy failures such as SSRF blocking remain HTTP errors.
#
# Configured `connector_id` plus `defer_loading: true` is validated
# here when the request also includes `tool_search`: the ID is resolved
# locally and stripped from the backend body, but this pipeline does
# **not** load tools. Deferred discovery requires `openai_mcp_dispatch`
# inside an agentic loop — see agentic-loop.yaml.
#
# On SUCCESS, the filter seeds one `mcp_list_tools` output item per
# resolved server into the response state, but — unlike the failure
# lifecycle, which this filter emits itself — a successful request must
# still reach the backend for inference, so those items are surfaced to
# the client only by a downstream response-finalizing filter:
# `openai_agentic_loop`/`openai_mcp_dispatch` (buffered `output`) or
# `openai_stream_events`, which synthesizes the streaming lifecycle
# when placed inside the agentic loop.
# This minimal pipeline has none of those, so it demonstrates resolution
# and the failure lifecycle only; see agentic-loop.yaml for a pipeline
# that surfaces successful `mcp_list_tools` items on both the buffered and
# streaming paths.
#
# This is a minimal pipeline focused on MCP tool discovery. For a
# complete setup with multi-turn conversation support (rehydrate,
# response store), see full-flow-agentic.yaml.
#
# For cross-request caching via `previous_tools`, add
# `openai_responses_rehydrate` before this filter so that
# `ResponsesState.previous_tools` is populated from the stored
# previous response.
#
# Example requests:
#
#   # Request with MCP tools
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{
#       "model": "gpt-4.1",
#       "input": "What is the weather?",
#       "tools": [{
#         "type": "mcp",
#         "server_label": "weather",
#         "server_url": "http://mcp-server:8001/mcp",
#         "allowed_tools": ["get_weather"]
#       }]
#     }'
#
#   # Request without MCP tools (passthrough)
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{"model": "gpt-4.1", "input": "Hello, world!"}'
#
#   # Request with connector_id (resolved from filter config)
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{
#       "model": "gpt-4.1",
#       "input": "Search for quarterly reports",
#       "tools": [{
#         "type": "mcp",
#         "server_label": "drive",
#         "connector_id": "corp_drive",
#         "allowed_tools": ["search"]
#       }]
#     }'
#
# Build:
#   cargo build -p praxis-ai-proxy --features openai-mcp-tools,store-sqlite

listeners:
  - name: ai-gateway
    address: "127.0.0.1:8080"
    filter_chains: [responses-pipeline]

filter_chains:
  - name: responses-pipeline
    filters:
      - filter: openai_responses_format
        on_invalid: continue
        headers:
          format: x-praxis-ai-format
          model: x-praxis-ai-model
          stream: x-praxis-ai-stream
          mode: x-praxis-responses-mode

      - filter: openai_tool_parse

      - filter: openai_mcp_tool_resolve
        timeout_ms: 5000
        connectors:
          - id: corp_drive
            server_url: https://drive-mcp.internal:8443/mcp
          - id: internal_search
            server_url: https://search-mcp.internal:8443/mcp

      - filter: openai_responses_proxy
        name: inference

      - filter: router
        routes:
          - path: "/v1/responses"
            headers:
              x-praxis-ai-format: "openai_responses"
            cluster: "inference-backend"

      - filter: load_balancer
        clusters:
          - name: "inference-backend"
            endpoints:
              - "127.0.0.1:3001"

insecure_options:
  allow_private_endpoints: true # example proxies to local backends