Plugin Lifecycle

Plugin Lifecycle

The plugin framework includes CLI tools to help you create, test, and deploy your plugins.

Development Flow

The plugin development workflow follows a straightforward workflow that gets you from idea to running plugin quickly.

graph LR

    A["template"]
    B(["$> bootstrap"])
    C(["$> build"])
    D(["$> serve"])

    subgraph dev
        A -.-> B
    end

    subgraph deploy
        C --> D
    end

    B --> C

    subgraph CF["Context Forge"]
        E["gateway"]
        D o--"MCP<br>&nbsp;&nbsp;<small>tools/call <i>hook</i></small>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;"--o E
    end

    style A stroke-dasharray: 3 3;

The process breaks down into these main steps:

• Bootstrap: Start with a plugin template and run the bootstrap command to set up your project structure
• Build: Compile and package your plugin code
• Serve: Launch a local development server to test your plugin
• Connect: Your served plugin automatically integrates with Context Forge's gateway via MCP (Model Context Protocol), using tool calls over standardized hooks for seamless communication

This streamlined approach lets developers focus on building functionality rather than wrestling with configuration, while ensuring plugins work smoothly within the broader Context Forge ecosystem.

Bootstrap

Creating a new plugin for Context Forge only takes a few minutes!

Using the mcpplugins tool (installed with Context Forge),

mcpplugins bootstrap --destination your/plugin/dir

The interactive prompt you guide you to enter plugin metadata, and will boostrap a complete plugin project for you including everything you need to kick the tires writing your new plugin.

For a full list of options, check:

mcpplugins bootstrap --help

!!! tip When prompted for the choosing the plugin type, select external to create standalone plugins (with their own lock files and dependency trees). Select native if you want to create a plugin that embeds and gets loaded directly into the gateway environment.

The examples under the plugins directory in the gateway repository serve as a guide of how to develop and test native plugins.

The following instructions apply to external plugins. First, change directory to work with your newly bootstrapped plugin:

cd your/plugin/dir
cp .env.template .env

Configuration

There are two main configuration files for your project.

./resources
    /plugins/config.yaml # configuration for your plugin and the plugin loader
    /runtime/config.yaml # configuration for the plugin server runtime

Inspect those two files and get familiar with them. There are many options you can customize, depending on your use case.

Dependencies

Plugins are Python packages with dependencies managed by uv. Just like the gateway, you can add, lock, lint, and ensure that best practices are followed when creating your plugins. To install dependencies with dev packages (required for linting and testing), run:

make install-dev

Alternatively, you can also install it in editable mode:

make install-editable

Test

To run all unit tests for your plugins:

make test

Build

To build a container image (runtime) containing a standardized plugin server, run:

make build

Serve

To start the plugin server:

make start

By default, this will start a Streamable HTTP MCP server on http://localhost:8000/mcp.

You can run mcp inspector to check your new server (note, it requires npm):

npx @modelcontextprotocol/inspector

Using gRPC Transport

For higher performance, you can use gRPC transport instead of MCP/HTTP. See the gRPC Transport Guide for details on performance comparisons.

Local Development with gRPC

# Install with gRPC support
make install-grpc

# Set transport to gRPC in .env
echo "PLUGINS_TRANSPORT=grpc" >> .env

# Start the server (defaults to port 50051)
./run-server.sh

Container Deployment with gRPC

To deploy your plugin container with gRPC transport:

# 1. Build the container with gRPC support
make build-grpc

# 2. Configure .env for gRPC
cat >> .env << 'EOF'
PLUGINS_TRANSPORT=grpc
PLUGINS_GRPC_SERVER_HOST=0.0.0.0
PLUGINS_GRPC_SERVER_PORT=50051
EOF

# 3. Start the container with gRPC port mapping
make start-grpc

Alternatively, you can manually specify the port:

make CONTAINER_PORT=50051 CONTAINER_INTERNAL_PORT=50051 start

Enabling mTLS for gRPC

To secure gRPC communication with mutual TLS:

# 1. Generate certificates (from the gateway directory)
make certs-mcp-ca                            # Generate CA
make certs-mcp-plugin PLUGIN_NAME=MyPlugin   # Generate plugin server cert

# 2. Copy certs to your plugin directory
mkdir -p certs/grpc
cp path/to/ca.pem certs/grpc/
cp path/to/server.pem certs/grpc/
cp path/to/server-key.pem certs/grpc/

# 3. Configure .env for mTLS
cat >> .env << 'EOF'
PLUGINS_TRANSPORT=grpc
PLUGINS_GRPC_SERVER_HOST=0.0.0.0
PLUGINS_GRPC_SERVER_PORT=50051
PLUGINS_GRPC_SERVER_SSL_ENABLED=true
PLUGINS_GRPC_SERVER_SSL_CERTFILE=certs/grpc/server.pem
PLUGINS_GRPC_SERVER_SSL_KEYFILE=certs/grpc/server-key.pem
PLUGINS_GRPC_SERVER_SSL_CA_CERTS=certs/grpc/ca.pem
PLUGINS_GRPC_SERVER_SSL_CLIENT_AUTH=require
EOF

# 4. Build and start with mTLS
make build-grpc
make start-grpc-tls   # Mounts certs/ into the container

The start-grpc-tls target automatically mounts the certs/ directory into the container at /opt/app-root/src/certs.

For detailed TLS/mTLS configuration options, see the gRPC Transport Guide.

Container Deployment with Unix Socket Transport

For local high-performance deployments using Unix domain sockets:

# 1. Build the container with gRPC support (required for protobuf)
make build-grpc

# 2. Configure .env for Unix socket
cat >> .env << 'EOF'
PLUGINS_TRANSPORT=unix
UNIX_SOCKET_PATH=/var/run/mcp-plugin.sock
EOF

# 3. Start with socket volume mount
docker run --name my-plugin \
    --env-file=.env \
    -v /var/run:/var/run \
    my-plugin:latest

Plugin Templates

The gateway ships with ready‑to‑use plugin templates under plugin_templates/ that the bootstrap tool uses. You can also copy them manually if you prefer.

Location and contents:

• plugin_templates/native

  • plugin.py.jinja: Plugin class skeleton extending Plugin
  • plugin-manifest.yaml.jinja: Manifest metadata (description, author, version, available_hooks)
  • config.yaml.jinja: Example entry to add in plugins/config.yaml
  • __init__.py.jinja, README.md.jinja

• plugin_templates/external

  • {{ plugin_name }}/plugin.py.jinja: External plugin implementation skeleton
  • resources/plugins/config.yaml.jinja: Plugin loader config for server
  • resources/runtime/config.yaml.jinja: External server runtime config
  • pyproject.toml.jinja, MANIFEST.in.jinja, .ruff.toml, .dockerignore, Containerfile
  • Makefile.jinja, run-server.sh: Build and serve helpers
  • tests/: Example tests and pytest config

Placeholders in filenames and files (e.g., {{ plugin_name.lower()... }}) are filled by the bootstrap tool. If copying manually, replace them with your plugin's name and details.

Using the templates via the CLI (recommended):

# Creates a new plugin from templates, asks interactive prompts
mcpplugins bootstrap --destination your/plugin/dir

# Pass --type to select template directly
mcpplugins bootstrap --destination your/plugin/dir --type native   # or external

After bootstrapping, follow the steps above to install deps, run tests, build, and serve.

Gateway Integration

Let's assume you have boostrapped the following plugin (resources/plugins/config.yaml) with default runtime (resources/runtime/config.yaml) options:

plugins:

  - name: "MyFilter"
    kind: "myfilter.plugin.MyFilter"
    description: "A filter plugin"
    version: "0.1.0"
    author: "Frederico Araujo"
    hooks: ["prompt_pre_fetch", "prompt_post_fetch", "tool_pre_invoke", "tool_post_invoke"]
    tags: ["plugin"]
    mode: "enforce"  # enforce | permissive | disabled
    priority: 150
    conditions:
      # Apply to specific tools/servers
      - server_ids: []  # Apply to all servers
        tenant_ids: []  # Apply to all tenants
    config:
      # Plugin config dict passed to the plugin constructor

# Plugin directories to scan
plugin_dirs:

  - "myfilter"

# Global plugin settings
plugin_settings:
  parallel_execution_within_band: true
  plugin_timeout: 30
  fail_on_plugin_error: false
  enable_plugin_api: true
  plugin_health_check_interval: 60

To integrate this plugin with the gateway, all you need to do is copying the following configuration under the plugins list in the gateway's plugins/config.yaml file:

plugins:
  # External Filter Plugin
  - name: "MyFilter"
    kind: "external"
    priority: 10 # adjust the priority
    mcp:
      proto: STREAMABLEHTTP
      url: http://localhost:8000/mcp

To use Streamable HTTP over a Unix domain socket (no TCP port):

```yaml
plugins:

  - name: "MyFilter"
    kind: "external"
    priority: 10
    mcp:
      proto: STREAMABLEHTTP
      url: http://localhost/mcp
      uds: /var/run/mcp-plugin.sock

To use STDIO instead of HTTP:

```yaml
plugins:

  - name: "MyFilter"
    kind: "external"
    priority: 10
    mcp:
      proto: STDIO
      cmd: ["python", "path/to/your/plugin_server.py"]
      env:
        PLUGINS_CONFIG_PATH: "/opt/plugins/config.yaml"
      cwd: "/opt/plugins"
      # Relative script paths are resolved from cwd when provided
      # or: script: path/to/your/plugin_server.py  # .py/.sh or executable

Then, start the gateway:

make serve

!!! note PLUGINS_ENABLED=true should be set in your gateway .env file.