feat: add mellea-mcp integration package by ajbozarth · Pull Request #54 · generative-computing/mellea-contribs

ajbozarth · 2026-05-04T22:54:51Z

Summary

Adds mellea-mcp, a new integration package that bridges MCP server tools into Mellea's native tool-calling system using the mcp SDK
Two-step public API: discover_mcp_tools() for inspection/filtering, MCPToolSpec.as_mellea_tool() for instantiation
Typed connection helpers: http_connection, sse_connection, stdio_connection
GitHub activity summary example demonstrating the full workflow with react()
24 unit tests across test_tools.py and test_connections.py

Test plan

uv sync --extra dev && uv run python -m pytest tests/ -v passes (24 tests)
uv run python examples/github_activity_summary.py runs end-to-end with a valid GITHUB_TOKEN

ajbozarth · 2026-05-04T22:57:34Z

I pinged everyone, but we should probably set up auto-assign reviewers on this repo like the others

akihikokuroda · 2026-05-04T23:25:52Z

Please move under mellea_contribs and follow https://github.com/generative-computing/mellea-contribs/blob/main/README.md#setting-up-a-new-subpackage-for-ci

Bridges MCP server tools into Mellea's native tool-calling system using the mcp SDK. Provides a two-step API: discover_mcp_tools() returns MCPToolSpec objects for inspection and filtering; as_mellea_tool() instantiates a callable MelleaTool. Includes typed connection helpers (http_connection, sse_connection, stdio_connection), a GitHub activity summary example using react(), and 24 unit tests. Signed-off-by: Alex Bozarth <ajbozart@us.ibm.com>

ajbozarth · 2026-05-05T00:23:53Z

Please move under mellea_contribs and follow https://github.com/generative-computing/mellea-contribs/blob/main/README.md#setting-up-a-new-subpackage-for-ci

Everything should be up to date now, I picked this work up from a very old branch and didn't rebase until after I opened the PR

ajbozarth · 2026-05-05T00:42:11Z

It's also worth noting that this was written with the intention of easily moving it into mellea if desired there instead

akihikokuroda

LGTM

psschwei

on first glance, seems mostly fine. will try to do a more thorough review later, but for now a couple of minor nits

- Tighten as_mellea_tool() return type from Any to MelleaTool - Replace per-call ThreadPoolExecutor with a module-level one to avoid thread creation overhead on every tool invocation - Handle EmbeddedResource text contents and surface binary content types (ImageContent, BlobResourceContents) as [binary: <mime>] descriptors rather than silently dropping them Signed-off-by: Alex Bozarth <ajbozart@us.ibm.com>

ajbozarth · 2026-05-05T17:40:56Z

@psschwei pushed an update addressing all you comments if you want to take another look

psschwei

there's a couple of things we may want to address, but I don't think any of them are blocking.

- Switch content block dispatch from hasattr to isinstance checks - Handle result.isError: surface error content as [tool error] prefix, falling back to a generic message when content is absent - Add tests for EmbeddedResource (text + blob), ImageContent, and isError cases; update existing sync wrapper tests to use real MCP types Signed-off-by: Alex Bozarth <ajbozart@us.ibm.com>

ajbozarth · 2026-05-05T20:17:28Z

@psschwei if you want to double check the fixes to your non-blockers before I merge this

Also @jakelorocco And @nrfulton I know you're super busy atm, but this contrib might be worth moving into mellea if you want to take a look next time you have a moment.

planetf1

Looks good generally - I just think there's some aspects of timeouts that need consideration

planetf1 · 2026-05-05T21:27:08Z

On example on ResourceLink being dropped - it's used by the github mcp server
Could be a followon pr if needed?

akihikokuroda · 2026-05-05T21:59:35Z

Would you update this line, too?

mellea-contribs/.github/workflows/build-package.yml

Line 89 in 17d2d0e

    
                     if [[ "$PACKAGE_DIR" == *"mellea-integration-core"* ]] || [[ "$PACKAGE_DIR" == *"reqlib_package"* ]] || [[ "$PACKAGE_DIR" == *"tools_package"* ]]; then

if this is expected cutting a release soon.

- Add configurable timeouts to all three transports: httpx.Timeout for streamable_http, timeout/sse_read_timeout for sse, asyncio.timeout for stdio - Expose connect_timeout/read_timeout on http_connection and sse_connection, and a single timeout param on stdio_connection - Handle AudioContent and ResourceLink content block types - Register atexit cleanup for the module-level thread pool executor - Fix _make_sync_call return type annotation to Callable[..., str] - Bump dependency floors to tested versions: mellea>=0.4.2, mcp>=1.27.0 - Add mcp_tools to skip_ollama allowlist in build-package.yml CI workflow Signed-off-by: Alex Bozarth <ajbozart@us.ibm.com>

ajbozarth · 2026-05-05T22:45:14Z

All review items should be addressed now

Falls back to a [resource: <uri>] placeholder if the read fails, since the MCP spec does not guarantee resource links are readable. Also adds AudioContent handling and tests for both new content types. Signed-off-by: Alex Bozarth <ajbozart@us.ibm.com>

jakelorocco

this looks nice; I will leave the final review for others. We should bring this up in the next Monday meeting that has everyone. I think it's a good candidate for being in Mellea itself, given that MCP servers are a common way to add tools to agents.

- Add mellea-mcp / mcp_tools entries to README.md, RELEASING.md, RELEASE_QUICK_START.md (package tables, directory tree, Ollama lists) - Tighten mcp_tools description to "Expose MCP server tools as Mellea tools" in pyproject.toml and propagate to docs - Remove README_mellea_contrib.md, an accidental near-duplicate of README.md introduced in generative-computing#47 - Drop stale "5 packages" counts in RELEASE_TESTING.md to avoid future drift when packages are added - Fix build-package.yml skip_ollama list: tools_package tests require Ollama (per ci.yml and test usage of start_session), so do not skip it at release time. Release path for mellea-tools has never been exercised upstream, so no prior release was affected. Signed-off-by: Alex Bozarth <ajbozart@us.ibm.com>

ajbozarth · 2026-05-06T17:18:38Z

Was looking at the repo page and realized I did not update any of the docs, so I just pushed an update to the docs, but while making that update I found a couple issues than I also addressed in c08be06:

There was a stale duplicate README committed by mistake
The README and CI were out of sync on tools_package so I fixed those (they were on the same lines I was already editing)

@akihikokuroda can you double check my changes?

psschwei · 2026-05-06T20:11:58Z

If we are going to bring into core (which I am on board with) maybe we don't merge this and then rip it right back out?

ajbozarth · 2026-05-06T20:14:08Z

If we are going to bring into core (which I am on board with) maybe we don't merge this and then rip it right back out?

That is fair, if my proposal to move to core would be implemented this sprint, but @serjikibm also made a comment in a recent call that this would be a great example of the contribs -> mellea pipeline

planetf1

Looks superb now -

planetf1 · 2026-05-07T08:07:37Z

Am ok with this being in core (but would re-review that pr in the mellea context)

psschwei · 2026-05-07T12:19:49Z

this would be a great example of the contribs -> mellea pipeline

I think that applies if it lived in contribs for a while, we saw it was useful, and decided that we should move it over to core. Since we've already decided to move it to core, that doesn't really apply here.

Now as you mentioned earlier, if it's going to take a while to get into core, then I could see having it here in the interim. I'd be inclined to at least see where the discussion goes Monday before making a decision (it's not like we're cutting a new version of contribs this week afaik).

jakelorocco · 2026-05-07T19:01:31Z

Discussed verbally; I am okay with this being in core. It's a smaller sized change that enables some good tool functionality. We should make sure to add some comments in the docs along side the current example though.

ajbozarth · 2026-05-08T00:40:14Z

Per team consensus I've opened generative-computing/mellea#1042 to add this to core instead of contribs and will close this.

github-actions Bot added the enhancement New feature or request label May 4, 2026

ajbozarth self-assigned this May 4, 2026

ajbozarth requested review from AngeloDanducci, akihikokuroda, planetf1 and psschwei May 4, 2026 22:56

ajbozarth force-pushed the mcp-integration branch 2 times, most recently from a8ca3db to 94ab69f Compare May 4, 2026 23:22

ajbozarth force-pushed the mcp-integration branch from 94ab69f to 57b9ea2 Compare May 5, 2026 00:22

akihikokuroda approved these changes May 5, 2026

View reviewed changes

psschwei reviewed May 5, 2026

View reviewed changes

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py Outdated

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py Outdated

psschwei approved these changes May 5, 2026

View reviewed changes

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py Outdated

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py

Comment thread mellea_contribs/mcp_tools/tests/test_tools.py

planetf1 requested changes May 5, 2026

View reviewed changes

planetf1 reviewed May 5, 2026

View reviewed changes

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py Outdated

planetf1 reviewed May 5, 2026

View reviewed changes

Comment thread mellea_contribs/mcp_tools/pyproject.toml Outdated

ajbozarth mentioned this pull request May 6, 2026

Team Challenge: Build Something You'll Actually Use, With Mellea generative-computing/mellea#1026

Closed

jakelorocco reviewed May 6, 2026

View reviewed changes

Comment thread mellea_contribs/mcp_tools/src/mellea_mcp/tools.py

ajbozarth requested a review from planetf1 May 6, 2026 15:03

ajbozarth mentioned this pull request May 6, 2026

Move MCP tool integration from contribs into core mellea generative-computing/mellea#1032

Closed

ajbozarth enabled auto-merge (squash) May 6, 2026 20:03

ajbozarth disabled auto-merge May 6, 2026 20:12

planetf1 approved these changes May 7, 2026

View reviewed changes

ajbozarth mentioned this pull request May 8, 2026

feat: add MCP tool integration generative-computing/mellea#1042

Merged

7 tasks

ajbozarth closed this May 8, 2026

ajbozarth deleted the mcp-integration branch May 8, 2026 00:40

Conversation

ajbozarth commented May 4, 2026

Summary

Test plan

Uh oh!

ajbozarth commented May 4, 2026

Uh oh!

akihikokuroda commented May 4, 2026

Uh oh!

ajbozarth commented May 5, 2026

Uh oh!

ajbozarth commented May 5, 2026

Uh oh!

akihikokuroda left a comment

Choose a reason for hiding this comment

Uh oh!

psschwei left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ajbozarth commented May 5, 2026

Uh oh!

psschwei left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ajbozarth commented May 5, 2026

Uh oh!

planetf1 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

planetf1 commented May 5, 2026

Uh oh!

akihikokuroda commented May 5, 2026

Uh oh!

ajbozarth commented May 5, 2026

Uh oh!

jakelorocco left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ajbozarth commented May 6, 2026

Uh oh!

psschwei commented May 6, 2026

Uh oh!

ajbozarth commented May 6, 2026

Uh oh!

planetf1 left a comment

Choose a reason for hiding this comment

Uh oh!

planetf1 commented May 7, 2026

Uh oh!

psschwei commented May 7, 2026

Uh oh!

jakelorocco commented May 7, 2026

Uh oh!

ajbozarth commented May 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants