feat(cluster): add MCP server to work with Platform (#218)

* feat(cluster): add MCP server to work with Platform

* refactor(mcp): inject client into mcp context

* refactor(mcp): use MCP to try and get around logging issues

* refactor(mcp): move cluster health into tool

* fix(MCP): use getattr not getitem to access client

* feat(MCP): add bot info resource, and start/stop bot

* fix(MCP): Catch ImportError for mcp usage

* feat(MCP): add more resources; add docs to existing tools/resources

* refactor(MCP): temporarily use tools instead of resources for stuff

* feat(MCP): add instructions; bot logs tool

* refactor(MCP): make MCP cluster-only, remove platform tools, add config

* feat(MCP): add variable group resources

* refactor(MCP): explicitly declare sse transport only

* feat(MCP): add a prompt for looking at cluster health

* refactor(MCP): change server name slightly

* feat(MCP): add `new_bot` and `remove_bot` tools

* docs(MCP): add notes to code why vargroups are not allowed C-UD methods

* docs(MCP): add some other docs behind design decisions

* feat(MCP): add `update_bot` tool

* refactor(MCP): only import mcp server if command is called

* refactor(CLI): hide display so that MCP server works w/ stdio transport

* refactor(MCP): update to using `fastmcp` because `mcp` module sucks

* refactor(MCP): drop resources, not a good fit

* feat(MCP): add tool to fetch URL of connected cluster

* refactor(MCP): show full bot log entry to see timestamps

* docs(MCP): add documentation for MCP CLI command & Cluster userguide

* style(CLI): fix line too long

* fix(Typing): forgot to install mcp extra for type checking

Author: fubuloubu

.github/workflows/test.yaml

@@ -52,7 +52,7 @@ jobs:
         - name: Install Dependencies
           run: |
             python -m pip install --upgrade pip
-            pip install .[lint,test]
+            pip install .[lint,test,mcp]
 
         - name: Run MyPy
           run: mypy .

docs/commands/cluster.rst

@@ -9,7 +9,7 @@ CLI commands for interacting with the Silverback Platform.
 .. click:: silverback._cli:cluster
     :prog: silverback cluster
     :nested: full
-    :commands: new, update, list, info, health
+    :commands: new, update, list, info, health, mcp
 
 .. click:: silverback._cli:workspaces
     :prog: silverback cluster workspaces
@@ -34,4 +34,4 @@ CLI commands for interacting with the Silverback Platform.
 .. click:: silverback._cli:pay
     :prog: silverback cluster pay
     :nested: full
-    :commands: create, add-time, cancel
+    :commands: create, add-time, cancel

docs/userguides/managing.md

@@ -45,8 +45,40 @@ Lastly, if you ever feel like you want to delete your Cluster, you can cancel th
 If you are the owner of the Stream, you can do this via the [`silverback cluster pay cancel`][silverback-cluster-pay-cancel] command.
 Only the owner may do this, so if you are not the owner you should contact them to have them do that action for you.
 
+## Model Context Server
+
+The Silverback package ships with an MCP ([Model Context Protocol](https://modelcontextprotocol.io/quickstart/user)) which you can use via the [`silverback cluster mcp`][silverback-cluster-mcp] command.
+This MCP server must be configured to run locally, and the easiest way to do so is to configure it in your LLM of choice.
+The config for using this with Claude Desktop is as follows:
+
+`~/.config/Claude/claude_desktop_config.json`:
+
+```json
+{
+  "mcpServers": {
+    ...  # Other MCP servers go here
+    "silverback": {
+      "command": "<path to `uvx` or `uv` binary>",
+      "args": [
+        "silverback[mcp]",
+        "cluster",
+        "mcp"
+        # Add args `--cluster <cluster name>` to use a Cluster other than your default
+      ]
+    }
+  }
+}
+```
+
+Once that has been configured, you can ask your LLM to do things like check the status of your cluster, summarize logs of running bots, and restart bots just be commanding it through a chat interface!
+
+```{notice}
+The MCP will use the context from [`silverback login`][silverback-login] to execute, so be sure to log in before starting.
+```
+
 [silverback-cluster-info]: ../commands/cluster.html#silverback-cluster-info
 [silverback-cluster-list]: ../commands/cluster.html#silverback-cluster-list
+[silverback-cluster-mcp]: ../commands/cluster.html#silverback-cluster-mcp
 [silverback-cluster-new]: ../commands/cluster.html#silverback-cluster-new
 [silverback-cluster-pay-add-time]: ../commands/cluster.html#silverback-cluster-pay-add-time
 [silverback-cluster-pay-cancel]: ../commands/cluster.html#silverback-cluster-pay-cancel

setup.py

@@ -3,6 +3,9 @@
 from setuptools import find_packages, setup
 
 extras_require = {
+    "mcp": [
+        "fastmcp",  # MCP Server & AI Computer Use
+    ],
     "test": [  # `test` GitHub Action jobs uses this
         "pytest>=6.0",  # Core testing package
         "pytest-xdist",  # Multi-process runner
@@ -44,6 +47,7 @@
     + extras_require["doc"]
     + extras_require["release"]
     + extras_require["dev"]
+    + extras_require["mcp"]
 )
 
 with open("./README.md") as readme: