Add Docker image build for MCP server

[nickmisasi]

Summary

• Adds CI workflow to build and push MCP server Docker image to Docker Hub on release tags
• Adds multi-stage Dockerfile for standalone MCP server container (linux/amd64 and linux/arm64)
• Adds make mcp-server-docker target for local builds
• Adds documentation for Docker-based deployment

Test plan

• Verify local build with make mcp-server-docker
• Test STDIO mode with Claude Code/Desktop
• Test HTTP mode connectivity

🤖 Generated with Claude Code

[github-actions]

🤖 LLM Evaluation Results

OpenAI

⚠️ Overall: 19/21 tests passed (90.5%)

Provider	Total	Passed	Failed	Pass Rate
⚠️ OPENAI	21	19	2	90.5%

❌ Failed Evaluations

Show 2 failures

OPENAI

1. TestChannelSummarization/[openai]_channel_summarization_developers_webapp_channel

• Score: 0.00
• Rubric: mentions @claudio.costa is working on adding code coverage tracking to the monorepo
• Reason: The output mentions @claudio.costa is tracking code coverage and sharing a PR, but it does not mention adding coverage tracking to the monorepo specifically.

2. TestChannelSummarization/[openai]_channel_summarization_developers_webapp_channel

• Score: 0.00
• Rubric: mentions claudio and harrison discussing exactly what should be tracked for code coverage
• Reason: While the output mentions Claudio and Harrison discussing coverage-related topics (snapshot tests and exploring E2E via Playwright) and a focus on the webapp, it does not state that they discussed exactly what should be tracked for code coverage. Therefore, it does not meet the rubric requirement.

---

Anthropic

⚠️ Overall: 13/14 tests passed (92.9%)

Provider	Total	Passed	Failed	Pass Rate
⚠️ ANTHROPIC	14	13	1	92.9%

❌ Failed Evaluations

Show 1 failures

ANTHROPIC

1. TestConversationMentionHandling/[anthropic]_conversation_from_attribution_long_thread.json

• Score: 0.00
• Rubric: is a list of bugs
• Reason: The output lists a mix of bugs, UX defects, and a feature gap explicitly labeled 'Not a Bug', so it is not strictly a list of bugs.

---

Azure OpenAI

✅ Overall: 21/21 tests passed (100.0%)

Provider	Total	Passed	Failed	Pass Rate
✅ AZURE	21	21	0	100.0%

---

Mistral

✅ Overall: 19/19 tests passed (100.0%)

Provider	Total	Passed	Failed	Pass Rate
✅ MISTRAL	19	19	0	100.0%

---

AWS Bedrock

✅ Overall: 25/25 tests passed (100.0%)

Provider	Total	Passed	Failed	Pass Rate
✅ BEDROCK	25	25	0	100.0%

---

This comment was automatically generated by the eval CI pipeline.

[NARSimoes]

I'm just adding a few points for discussion - overall this looks good.

[NARSimoes]

I'd consider using commit SHA pointing to a specific commit for better reproducibility. For example, instead of actions/[email protected] we can use something like actions/checkout@ 8e8c483db84b4bee98b60c0593521ed34d9990e8 #v6.0.1. The same applies for there steps.

[NARSimoes]

What do you think about adding a step to scan the Dockerfile & images on pull-request and before pushing to the repository. With that we can catch issues earlier in CI and before those images are published.

[NARSimoes]

What do you think about breaking this down to also validate the Dockerfile in pull-requests ? For example, something like:

• On pull-requests: checkout -> build -> scan / lint / test Dockerfile & images
• On github.ref, 'refs/tags/v': push the images