bld MCP
A Java build tool with an MCP mode: an agent compiles, tests and runs the project through its own build commands
- 323 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent runs project build commands, which execute arbitrary Java code and tests on your machine
- If the build defines a publishing task, the agent could push artifacts to Maven Central
Install
- Works in
Manual install
./bld mcp installRun in the root of a bld project; writes .mcp.json.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs project build commands, which execute arbitrary Java code and tests on your machine
- If the build defines a publishing task, the agent could push artifacts to Maven Central
This is third-party code. Review the repository files before installing.
What it does
bld is a build system where build logic is written in plain Java. Since version 3.0 it can serve its build commands to an agent as MCP tools. The agent gets a structured interface for compiling, testing and running the project. Command output arrives as a stream with execution metadata, so there is no terminal text to parse. Connecting takes one command in the project root. `./bld mcp install` writes .mcp.json for Claude Code and compatible clients, and with the cursor or vscode argument it writes the config for Cursor or VS Code. With the `--print` flag the configuration is printed and nothing is written. If the file already exists, bld adds its entry and keeps the other servers. The server itself starts with `./bld mcp` and runs over stdio until the input stream closes. The registered command launches the project wrapper directly through java, so the same file works on every platform. Build commands stay explicit, as everywhere in bld: nothing happens until you or the agent call a task. Beyond building, bld resolves dependencies from Maven Central and builds images with jlink and jpackage. Artifacts are published through the Sonatype Central Portal.
Who it is for. For Java developers who build with bld and want Claude Code or Cursor to run builds and tests without parsing text output.
Good fit when
- The project already builds with bld and the agent should compile code and run tests on its own
- You want streamed output and execution metadata with no terminal-text parsing
- The same MCP config must work on Windows, macOS and Linux in a shared team repository
- You want to connect the build to Claude Code, Cursor and VS Code at once
Not a fit when
- The project builds with Maven or Gradle: the MCP mode works only in bld projects
- You need stable functionality: the MCP mode is still experimental and may change
- Your bld version is below 3.0: the server mode arrived in 3.0
- The machine has no Java 17 or newer
Example request
Build the project with bld, run the tests and show the failing ones with the error text.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The bld help calls the MCP mode experimental: the tool set and behavior may change. Java 17 or newer and a project already set up for bld are required.
The server exposes only the build commands defined in the project. If the build publishes artifacts or runs arbitrary code, the agent gets the same abilities. The help does not list exact tool names, so check them in your client's tool list before relying on them.
How to disable. Remove the bld entry from .mcp.json, .cursor/mcp.json or .vscode/mcp.json, depending on your client. The build tool itself keeps working as before.
FAQ
How do I connect bld to Claude Code, Cursor or VS Code?
In the project root run `./bld mcp install` for Claude Code, `./bld mcp install cursor` for Cursor or `./bld mcp install vscode` for VS Code. After a restart the client lists the build tools.
Which bld version has the MCP mode?
Version 3.0. Older versions have no `./bld mcp` command, so update the build tool first.
What does the --print flag do?
`./bld mcp install --print` prints the configuration to the screen and writes nothing to disk. It lets you check the entry before installing.
How is bld different from Maven and Gradle?
Maven describes a build in XML and Gradle in Groovy or Kotlin, while bld is ordinary Java code. The IDE gives completion, and the debugger works on the build itself. bld ships as a single jar.
Which transport does the server use?
Standard input and output using the MCP stdio transport. The server runs until the input stream closes.
Discussion
No questions or reviews yet.
Related
MCP serversA lightweight CLI to discover, inspect and call MCP server tools straight from the shell
MCP wrapper for BSL Language Server
BSL MCP Server
A small MCP server that runs BSL Language Server as a JAR file and gives an agent BSL code analysis and formatting as JSON
A script installs BSL Language Server as a Linux systemd service and, with a flag, opens an MCP endpoint for AI agents
MCP bridge to BSL Language Server: navigation, diagnostics and refactoring of 1C and OneScript code for agents
An MCP server that lets an agent read files, search code and run commands in a workspace
An MCP server that exposes the cloud 1C:Napparnik (1C.ai) assistant and a local BSL Language Server to agents in Cursor and other IDEs
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| bld (this entry) | yes | not required | 323 | |
| Plinth | no | 456 | skills and agents for Java development: build, tests, profiling and architecture, without a build system of its own | |
| claude-code-java | no | 761 | a set of skills for Java projects: review, Spring Boot, JPA and Maven audits |
README in short
The project presents itself as a build tool for people who dislike fussing with builds: tasks never run unasked and there is no magic. Next to the MCP mode there are extensions for Kotlin, Spring Boot and Checkstyle and an IntelliJ IDEA plugin.