Move `ldk-server-mcp` into the workspace by tnull · Pull Request #188 · lightningdevkit/ldk-server

tnull · 2026-04-15T09:37:11Z

Moving from https://github.com/tnull/ldk-server-mcp

Draft for now.

ldk-reviews-bot · 2026-04-15T09:37:14Z

🎉 This PR is now ready for review!
Please choose at least one reviewer by assigning them on the right bar.
If no reviewers are assigned within 10 minutes, I'll automatically assign one.
Once the first reviewer has submitted a review, a second will be assigned if required.

benthecarman

Did a quick skim and ran through codex

Codex Review:

  - [P2] Honor storage.disk.dir_path when resolving credentials — /home/ben/projects/ldk-server/ldk-server-mcp/src/
    config.rs:57-60
    If the shared ldk-server config uses a custom storage.disk.dir_path, this parser drops that section entirely, so
    resolve_config() can only look in ~/.ldk-server/... for api_key and tls.crt. In deployments that store node state
    outside the default directory (a setup ldk-server-cli already supports), the MCP bridge will fail to authenticate
    unless users also set both LDK_API_KEY and LDK_TLS_CERT_PATH by hand.
  - [P2] Fall back to the default gRPC address when no base URL is configured — /home/ben/projects/ldk-server/ldk-
    server-mcp/src/config.rs:112-116
    In setups that rely on the standard 127.0.0.1:3536 gRPC address and only use the default credential files on disk,
    this path exits with “Base URL not provided” instead of using the same default as ldk-server-cli. As a result, ldk-
    server-mcp cannot start in the zero-config case unless users create a config file or export LDK_BASE_URL, even
    though the server is running at the normal address.

I think these are two things we fixed in the cli since you created the MCP server.

benthecarman

needs rebase too

benthecarman · 2026-04-26T22:46:34Z

+// string, so we manually extract the string and turn it into bytes instead of delegating to the
+// proto-typed deserializer.
+pub async fn handle_sign_message(client: &LdkServerClient, args: Value) -> Result<Value, McpError> {
+	let message = args


any reason we aren't using the json decoder here? and in handle_verify_signature

Hmm, see the comment above. Now introduced some extra {SignMessage,VerifySignature}Args structs deriving Deserialize to make this a bit cleaner. Let me know what you think.

benthecarman · 2026-04-26T22:49:30Z

+- **Spec**: https://spec.modelcontextprotocol.io/
+- **Transport**: stdio (one JSON-RPC 2.0 message per line)
+- **Methods implemented**: `initialize`, `tools/list`, `tools/call`
+- **Notifications handled**: `notifications/initialized` (ignored, no response)


I don't see us handling this anywhere, looks like we'd return METHOD_NOT_FOUND

No, we just skip in main:

// Notifications have no id — do not respond if request.id.is_none() { continue; }

+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v6
+      - uses: dtolnay/rust-toolchain@nightly


+      - name: Get the current date
+        run: echo "date=$(date +'%Y-%m-%d')" >> $GITHUB_ENV
+      - name: Create Pull Request
+        uses: peter-evans/create-pull-request@v8


+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v6
+      - uses: dtolnay/rust-toolchain@nightly


+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v6
+      - uses: dtolnay/rust-toolchain@nightly


+      - name: Get the current date
+        run: echo "date=$(date +'%Y-%m-%d')" >> $GITHUB_ENV
+      - name: Create Pull Request
+        uses: peter-evans/create-pull-request@v8


Derive `serde::Deserialize` (alongside the existing `serde::Serialize`) and `#[serde(default)]` on all generated prost messages so that JSON payloads can be deserialized directly into the request types. This reinstates the deserializers that were dropped in 8af3189 so that clients like `ldk-server-mcp` can parse tool arguments straight into the typed proto structs instead of threading every field through `serde_json::Value`. Generated with the assistance of AI (Claude).

benthecarman · 2026-05-01T22:29:59Z

Can you squash?

tnull · 2026-05-05T09:07:00Z

Can you squash?

Squashed without further changes.

benthecarman

final thing and then i think this is good

benthecarman · 2026-05-05T17:02:53Z

+	if request.expiry_secs == 0 {
+		request.expiry_secs = DEFAULT_EXPIRY_SECS;


I think we'll need to do similar logic for DEFAULT_MAX_TOTAL_CLTV_EXPIRY_DELTA, DEFAULT_MAX_PATH_COUNT, and DEFAULT_MAX_CHANNEL_SATURATION_POWER_OF_HALF where relevant

benthecarman

lgtm on squash!

Move the MCP bridge into the ldk-server workspace and switch it to an in-tree client dependency so workspace builds and tests cover it directly. Co-Authored-By: HAL 9000

Use the workspace-built MCP binary directly in tests so regressions are caught without requiring a live server and without recompiling from inside each test. Generated with the assistance of AI (Claude).

Build the MCP binary in the e2e harness and exercise its stdio protocol against a live ldk-server so basic MCP functionality is verified without involving an agent. Co-Authored-By: HAL 9000

Run MCP-specific formatting, clippy, and crate tests in a dedicated workflow and add a separate job that exercises the MCP e2e sanity suite on Ubuntu. Co-Authored-By: HAL 9000

Describe the MCP bridge as a first-class workspace member and document how to build, test, and sanity-check it alongside the rest of ldk-server. Co-Authored-By: HAL 9000

tnull · 2026-05-07T07:56:34Z

lgtm on squash!

Squashed without further changes.

tnull requested a review from benthecarman April 15, 2026 09:37

tnull marked this pull request as draft April 15, 2026 09:37

tnull force-pushed the 2026-04-move-mcp-into-workspace branch 2 times, most recently from 70f8e5f to 3bc8a6e Compare April 15, 2026 09:50

benthecarman reviewed Apr 15, 2026

View reviewed changes

Comment thread .github/workflows/mcp.yml Outdated

Comment thread .github/workflows/mcp.yml Outdated

Comment thread e2e-tests/build.rs Outdated

Comment thread ldk-server-mcp/src/tools/handlers.rs Outdated

Comment thread ldk-server-mcp/README.md

tnull force-pushed the 2026-04-move-mcp-into-workspace branch from 3bc8a6e to 8d6d7b0 Compare April 16, 2026 09:47

tnull requested a review from benthecarman April 16, 2026 09:47

tnull mentioned this pull request Apr 16, 2026

Add initial guides & documentation #186

Merged

tnull marked this pull request as ready for review April 16, 2026 11:03

benthecarman reviewed Apr 16, 2026

View reviewed changes

tnull force-pushed the 2026-04-move-mcp-into-workspace branch 2 times, most recently from b76ffeb to b0292ff Compare April 23, 2026 13:37

benthecarman reviewed Apr 26, 2026

View reviewed changes

tnull force-pushed the 2026-04-move-mcp-into-workspace branch from b0292ff to 9295c38 Compare April 27, 2026 13:41

github-advanced-security AI found potential problems Apr 27, 2026

View reviewed changes

tnull force-pushed the 2026-04-move-mcp-into-workspace branch from 9295c38 to 9d74698 Compare April 27, 2026 13:44

tnull force-pushed the 2026-04-move-mcp-into-workspace branch from 9d74698 to 0ff77b7 Compare May 5, 2026 09:05

benthecarman reviewed May 5, 2026

View reviewed changes

tnull force-pushed the 2026-04-move-mcp-into-workspace branch from acfab96 to e729511 Compare May 6, 2026 12:29

benthecarman reviewed May 6, 2026

View reviewed changes

tnull added 5 commits May 7, 2026 09:50

Add ldk-server-mcp to the workspace

1b48e36

Move the MCP bridge into the ldk-server workspace and switch it to an in-tree client dependency so workspace builds and tests cover it directly. Co-Authored-By: HAL 9000

Improve mcp test coverage in workspace

ae179b4

Use the workspace-built MCP binary directly in tests so regressions are caught without requiring a live server and without recompiling from inside each test. Generated with the assistance of AI (Claude).

Add end-to-end sanity checks for ldk-server-mcp

35eeaf4

Build the MCP binary in the e2e harness and exercise its stdio protocol against a live ldk-server so basic MCP functionality is verified without involving an agent. Co-Authored-By: HAL 9000

Add CI coverage for ldk-server-mcp

ae6e3f2

Run MCP-specific formatting, clippy, and crate tests in a dedicated workflow and add a separate job that exercises the MCP e2e sanity suite on Ubuntu. Co-Authored-By: HAL 9000

Document ldk-server-mcp in the workspace

429b1a7

Describe the MCP bridge as a first-class workspace member and document how to build, test, and sanity-check it alongside the rest of ldk-server. Co-Authored-By: HAL 9000

tnull force-pushed the 2026-04-move-mcp-into-workspace branch from e729511 to 429b1a7 Compare May 7, 2026 07:51

Anyitechs mentioned this pull request May 7, 2026

Add ldk-server-mcp crate scaffolding for MCP gateway #202

Closed

6 tasks

		if request.expiry_secs == 0 {
		request.expiry_secs = DEFAULT_EXPIRY_SECS;

Conversation

tnull commented Apr 15, 2026

Uh oh!

ldk-reviews-bot commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

benthecarman left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

benthecarman 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!

Uh oh!

Uh oh!

benthecarman Apr 26, 2026

Choose a reason for hiding this comment

Uh oh!

tnull Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

benthecarman Apr 26, 2026

Choose a reason for hiding this comment

Uh oh!

tnull Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

benthecarman commented May 1, 2026

Uh oh!

tnull commented May 5, 2026

Uh oh!

benthecarman left a comment

Choose a reason for hiding this comment

Uh oh!

benthecarman May 5, 2026

Choose a reason for hiding this comment

Uh oh!

benthecarman left a comment

Choose a reason for hiding this comment

Uh oh!

tnull commented May 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

ldk-reviews-bot commented Apr 15, 2026 •

edited

Loading