Agent Gateway: resolve_image + deploy_ephemeral MCP tools (standalone scalable server)

.env.agent-gateway.example

@@ -0,0 +1,37 @@
+# Example env for docker-compose.agent-gateway.yml
+#   cp .env.agent-gateway.example .env
+#   docker compose -f docker-compose.agent-gateway.yml up
+#
+# Docker Compose auto-loads a file named `.env` next to the compose file for
+# ${VAR} substitution. Every value below has a sane default baked into the
+# compose, so an empty/absent .env still works — override only what you need.
+
+# --- Which image to run ------------------------------------------------------
+# The tag this branch's CI publishes. Use a different tag once merged to dev/main.
+AGENT_GATEWAY_TAG=feature-agent-gateway
+
+# Host port to expose the gateway on (container always listens on 4600).
+HOST_PORT=4600
+
+# --- Reusing the running TryDirect stack ------------------------------------
+# The compose joins the external trydirect networks and talks to the existing
+# `stackerdb`/`stackeredis`. The gateway itself reads DB creds from the baked
+# configuration.yaml (host=stackerdb, db=stacker, postgres/postgres). Only the
+# one-shot `migrate` uses DATABASE_URL — override it if your stackerdb differs
+# (and mount a matching configuration.yaml for the gateway; see the compose).
+DATABASE_URL=postgres://postgres:postgres@stackerdb:5432/stacker
+
+# --- Sandbox quotas (deploy_ephemeral only) ---------------------------------
+# These gate deploy_ephemeral's safety/quota checks. Provisioning itself is M3
+# (needs RabbitMQ/Vault/install-service), so these don't spend money yet.
+SANDBOX_DEFAULT_TTL_SECS=1800     # 30m
+SANDBOX_MAX_TTL_SECS=7200         # 2h
+SANDBOX_MAX_CONCURRENT_PER_USER=3
+
+# --- What actually works in this test image ---------------------------------
+# GET  /health
+# POST /public/resolve_image   {"reference":"redis:7-alpine"}   (no auth, real data)
+# WS   /mcp                     (needs a user JWT; resolve_image is anonymous there too)
+#
+# Postgres creds are fixed to match the image's baked configuration.yaml
+# (host=stackerdb, db=stacker, postgres/postgres) — don't change them here.

.github/workflows/agent-gateway.yml

@@ -0,0 +1,83 @@
+name: Agent Gateway
+
+on:
+  workflow_dispatch: {}
+  push:
+    branches: [dev, main, feature/agent-gateway]
+    paths:
+      - "crates/agent-tools/**"
+      - "crates/td-audit/**"
+      - "src/mcp/**"
+      - "src/routes/agent_public.rs"
+      - "migrations/**"
+      - "src/bin/agent_gateway.rs"
+      - "src/startup.rs"
+      - "Cargo.toml"
+      - "Cargo.lock"
+      - "Dockerfile.agent-gateway"
+      - ".github/workflows/agent-gateway.yml"
+  pull_request:
+    branches: [dev, main]
+    paths:
+      - "crates/agent-tools/**"
+      - "crates/td-audit/**"
+      - "src/mcp/**"
+      - "src/routes/agent_public.rs"
+      - "migrations/**"
+      - "src/bin/agent_gateway.rs"
+      - "src/startup.rs"
+      - "Dockerfile.agent-gateway"
+      - ".github/workflows/agent-gateway.yml"
+
+jobs:
+  # Fast, DB-less feedback: the pure crate + that the standalone binary compiles.
+  test:
+    name: Test crate + build binary
+    runs-on: ubuntu-latest
+    env:
+      SQLX_OFFLINE: true
+    steps:
+      - uses: actions/checkout@v4
+      - name: Install protoc
+        run: sudo apt-get update -qq && sudo apt-get install -y protobuf-compiler
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+      - name: Cache cargo
+        uses: Swatinem/rust-cache@v2
+      - name: Unit-test the pure agent-tools crate (no DB/cloud)
+        run: cargo test -p agent-tools --verbose
+      - name: Build the agent-gateway binary (offline)
+        run: cargo build --bin agent-gateway --verbose
+
+  # Build + push the image on push to dev/main/feature (not PRs). The Docker
+  # build compiles inside rust:bookworm, so glibc matches the runtime — no
+  # coupling to the runner's glibc.
+  docker:
+    name: Build & push image
+    needs: test
+    if: github.event_name == 'push' || github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKER_USERNAME }}
+          password: ${{ secrets.DOCKER_PASSWORD }}
+      - name: Set image tag
+        id: tags
+        # Sanitize the ref for a valid Docker tag (feature/x -> feature-x).
+        run: |
+          TAG="$(echo "${{ github.ref_name }}" | tr '/' '-')"
+          echo "tags=trydirect/agent-gateway:${TAG}" >> "$GITHUB_OUTPUT"
+      - name: Build and push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./Dockerfile.agent-gateway
+          push: true
+          tags: ${{ steps.tags.outputs.tags }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.github/workflows/rust.yml

@@ -62,12 +62,16 @@ jobs:
       - name: Build stacker-cli (release)
         run: cargo build --release --target ${{ matrix.target }} --bin stacker-cli --verbose
 
+      - name: Build agent-gateway (release)
+        run: cargo build --release --target ${{ matrix.target }} --bin agent-gateway --verbose
+
       - name: Prepare binaries
         run: |
           mkdir -p artifacts
           cp target/${{ matrix.target }}/release/server artifacts/server
           cp target/${{ matrix.target }}/release/console artifacts/console
           cp target/${{ matrix.target }}/release/stacker-cli artifacts/stacker-cli
+          cp target/${{ matrix.target }}/release/agent-gateway artifacts/agent-gateway
           tar -czf ${{ matrix.artifact_name }}.tar.gz -C artifacts .
       - name: Upload binaries
         uses: actions/upload-artifact@v4

Cargo.toml

@@ -5,7 +5,7 @@ edition = "2021"
 default-run= "server"
 
 [workspace]
-members = ["crates/pipe-adapter-sdk", "crates/pipe-adapter-mail", "crates/td-audit"]
+members = ["crates/pipe-adapter-sdk", "crates/pipe-adapter-mail", "crates/td-audit", "crates/agent-tools"]
 resolver = "2"
 
 [lib]
@@ -28,6 +28,10 @@ name = "stacker-cli"
 path = "src/bin/agent_executor.rs"
 name = "agent-executor"
 
+[[bin]]
+path = "src/bin/agent_gateway.rs"
+name = "agent-gateway"
+
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
@@ -39,6 +43,7 @@ config = "0.13.4"
 reqwest = { version = "0.11.23", features = ["json", "blocking", "stream", "native-tls"] }
 serde = { version = "1.0.195", features = ["derive"] }
 td-audit = { path = "crates/td-audit" }
+agent-tools = { path = "crates/agent-tools" }
 tokio = { version = "1.28.1", features = ["full"] }
 tracing = { version = "0.1.40", features = ["log"] }
 tracing-bunyan-formatter = "0.3.8"

Dockerfile.agent-gateway

@@ -0,0 +1,56 @@
+# syntax=docker/dockerfile:1.4
+# Standalone image for the agent-gateway MCP server.
+#
+# Multi-stage: compile INSIDE rust:bookworm and run on debian:bookworm-slim, so
+# the binary's glibc always matches the runtime (no coupling to the CI runner's
+# glibc — which caused "GLIBC_2.38 not found" when baking an ubuntu-built binary).
+FROM rust:bookworm AS builder
+
+RUN apt-get update && apt-get install --no-install-recommends -y \
+        protobuf-compiler libprotobuf-dev libssl-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN cargo install sqlx-cli --no-default-features --features native-tls,postgres
+
+WORKDIR /app
+
+# Manifests + offline sqlx cache (compile-time checked queries without a DB).
+COPY ./Cargo.toml .
+COPY ./Cargo.lock .
+COPY ./build.rs .
+COPY ./rustfmt.toml .
+COPY ./docker/local/.env .
+COPY ./docker/local/configuration.yaml .
+COPY .sqlx .sqlx/
+COPY ./proto ./proto
+COPY ./tests/bdd.rs ./tests/bdd.rs
+
+COPY ./src ./src
+COPY ./crates ./crates
+COPY ./scenarios ./scenarios
+
+ENV SQLX_OFFLINE=true
+RUN cargo build --release --bin agent-gateway
+
+# ---- runtime ----
+FROM debian:bookworm-slim AS production
+
+RUN apt-get update && apt-get install --no-install-recommends -y \
+        libssl3 ca-certificates \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+RUN mkdir ./files && chmod 0777 ./files
+
+COPY --from=builder /app/target/release/agent-gateway /app/agent-gateway
+COPY --from=builder /usr/local/cargo/bin/sqlx /usr/local/bin/sqlx
+COPY docker/local/configuration.yaml /app/configuration.yaml
+COPY docker/local/.env /app/.env
+COPY access_control.conf.dist /app/access_control.conf
+# Baked migrations so `sqlx migrate run` works from the image.
+COPY migrations /app/migrations
+
+ENV AGENT_GATEWAY_BIND=0.0.0.0:4600
+EXPOSE 4600
+
+ENTRYPOINT ["/app/agent-gateway"]

crates/agent-tools/Cargo.toml

@@ -0,0 +1,16 @@
+[package]
+name = "agent-tools"
+version = "0.1.0"
+edition = "2021"
+description = "Pure-logic core for TryDirect's agent-facing MCP tools (resolve_image ground truth, deploy_ephemeral sandbox orchestration). Heavy I/O is injected via traits so this compiles/tests apart from the Actix/sqlx server."
+
+[dependencies]
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"
+serde_yaml = "0.9"
+thiserror = "1"
+async-trait = "0.1"
+td-audit = { path = "../td-audit" }
+
+[dev-dependencies]
+tokio = { version = "1", features = ["macros", "rt-multi-thread"] }

crates/agent-tools/src/error.rs

@@ -0,0 +1,24 @@
+//! Error type shared by the agent tools. Kept small and stringly-mappable so MCP
+//! `ToolHandler`s can turn it into a JSON-RPC error message.
+
+use thiserror::Error;
+
+#[derive(Debug, Error, PartialEq)]
+pub enum AgentToolError {
+    #[error("invalid input: {0}")]
+    InvalidInput(String),
+
+    #[error("quota exceeded: {0}")]
+    QuotaExceeded(String),
+
+    /// The agent-supplied compose was rejected by the safety gate before any
+    /// infrastructure was provisioned.
+    #[error("compose rejected: {0}")]
+    ComposeRejected(String),
+
+    /// A downstream (registry, MQ, cloud, DB) operation failed.
+    #[error("backend error: {0}")]
+    Backend(String),
+}
+
+pub type Result<T> = std::result::Result<T, AgentToolError>;