Skip to content

Automate browser based workflows with AI

chore: add Glama MCP Dockerfile and release docs (#5418)

Marc Kelechavab7ba2af57e54
diff --git a/integrations/mcp/Dockerfile b/integrations/mcp/Dockerfile
new file mode 100644
index 000000000..09362371f
--- /dev/null
+++ b/integrations/mcp/Dockerfile
@@ -0,0 +1,51 @@
+FROM python:3.11 AS requirements-stage
+
+WORKDIR /tmp
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh && \
+ ln -s /root/.local/bin/uv /usr/local/bin/uv
+COPY pyproject.toml /tmp/pyproject.toml
+COPY uv.lock /tmp/uv.lock
+RUN uv pip compile pyproject.toml -o requirements.txt --no-annotate --no-header
+
+FROM python:3.11-slim-bookworm
+
+ENV PYTHONUNBUFFERED=1 \
+ PIP_NO_CACHE_DIR=1 \
+ SKYVERN_API_KEY="" \
+ SKYVERN_MCP_TRANSPORT=stdio \
+ SKYVERN_MCP_PATH=/mcp \
+ PORT=8000
+
+WORKDIR /app
+
+COPY --from=requirements-stage /tmp/requirements.txt /app/requirements.txt
+
+RUN apt-get update && \
+ apt-get install -y --no-install-recommends ca-certificates curl gnupg && \
+ rm -rf /var/lib/apt/lists/*
+
+COPY .nvmrc /app/.nvmrc
+COPY nodesource-repo.gpg.key /tmp/nodesource-repo.gpg.key
+RUN mkdir -p /etc/apt/keyrings && \
+ cat /tmp/nodesource-repo.gpg.key | gpg --dearmor -o /etc/apt/keyrings/nodesource.gpg && \
+ NODE_MAJOR="$(cut -d. -f1 < /app/.nvmrc)" && \
+ echo "deb [signed-by=/etc/apt/keyrings/nodesource.gpg] https://deb.nodesource.com/node_${NODE_MAJOR}.x nodistro main" \
+ > /etc/apt/sources.list.d/nodesource.list && \
+ apt-get update && \
+ apt-get install -y --no-install-recommends nodejs && \
+ npm install -g @bitwarden/[email protected] && \
+ rm -rf /var/lib/apt/lists/* /tmp/nodesource-repo.gpg.key
+
+RUN python -m pip install --upgrade pip setuptools wheel && \
+ python -m pip install -r /app/requirements.txt && \
+ playwright install --with-deps chromium
+
+COPY . /app
+
+ENV PYTHONPATH=/app
+
+EXPOSE 8000
+
+# Default to stdio for Glama inspection. If Glama is configured to run the
+# container over HTTP, set SKYVERN_MCP_TRANSPORT=streamable-http (or sse).
+CMD ["sh", "-lc", "if [ \"${SKYVERN_MCP_TRANSPORT}\" = \"stdio\" ]; then exec python -m skyvern run mcp; else exec python -m skyvern run mcp --transport \"${SKYVERN_MCP_TRANSPORT}\" --host 0.0.0.0 --port \"${PORT}\" --path \"${SKYVERN_MCP_PATH}\"; fi"]
diff --git a/integrations/mcp/README.md b/integrations/mcp/README.md
index 7faddaac7..9a45e9cb5 100644
--- a/integrations/mcp/README.md
+++ b/integrations/mcp/README.md
@@ -76,3 +76,34 @@ Use the following config if you want to set up Skyvern for any other MCP-enabled
}
}
```
+
+## Glama Release Setup
+
+Glama's "release" flow is different from publishing the package to PyPI or the
+official MCP Registry. For Glama, you need a runnable server container so Glama
+can boot the MCP server, inspect the tool schema, and publish an installable
+release in their directory.
+
+Use the dedicated [`Dockerfile`](./Dockerfile) in this directory for that flow.
+The root [`Dockerfile`](../../Dockerfile) is for the full Skyvern app stack and
+starts `python -m skyvern.forge`, which is the wrong runtime for an MCP-only
+Glama release.
+
+Recommended Glama setup:
+
+1. Claim the server in Glama. This repository already includes
+ [`glama.json`](../../glama.json), so authorized maintainers can claim the
+ `Skyvern-AI/skyvern` entry.
+2. In Glama's Dockerfile admin page, point the build to `Dockerfile.glama`.
+3. Keep the default command unless Glama explicitly asks for HTTP transport.
+ The image defaults to `python -m skyvern run mcp` over stdio.
+4. If you want the hosted Glama release to use Skyvern Cloud browser sessions,
+ add a real `SKYVERN_API_KEY` secret in Glama. Otherwise the container boots
+ in local embedded mode, which is enough for inspection but not ideal for
+ cloud-backed browser sessions.
+5. Deploy, wait for inspection to pass, then use Glama's "Make Release" action
+ in the server admin UI.
+
+If you are also publishing to the official MCP Registry, treat that as a
+separate step. The official registry uses package metadata and `server.json`;
+Glama releases are container-based.