mirror of
https://github.com/DuendeSoftware/products
synced 2026-05-24 09:28:24 +00:00
Docs MCP server - Update README
This commit is contained in:
Maarten Balliauw
parent
8189f9f5a9
commit
3464b40468
1 changed files with 89 additions and 26 deletions
|
|
@ -1,38 +1,95 @@
|
|||
# Duende.Documentation.Mcp.Server
|
||||
# Duende Documentation MCP Server
|
||||
|
||||
This project contains an MCP server that can extend any large language model that supports MCP with up-to-date knowledge about Duende products.
|
||||
MCP is an open protocol that enables AI models to securely interact with local and remote resources through standardized
|
||||
server implementations. This project contains an MCP server that can extend any large language model that supports MCP
|
||||
with up-to-date knowledge about Duende products, sourced from [documentation](https://docs.duendesoftware.com/),
|
||||
[blog](https://duendesoftware.com/blog/) and [samples](https://github.com/duendesoftware/samples).
|
||||
|
||||
## MCP Tools
|
||||
## Register the MCP
|
||||
|
||||
The server has several tools available:
|
||||
* Free text search on blogs, docs, or samples
|
||||
* Fetch specific page
|
||||
* Get all content for a sample
|
||||
* Get a specific file from a sample
|
||||
To run the Duende Documentation MCP Server, you will need the `dnx` tool (included in the .NET 10 SDK) in your
|
||||
system's `PATH`. The `dnx` tool can download and run applications packaged and distributed through NuGet.
|
||||
|
||||
The MCP server has [instructions](src/Duende.Documentation.Mcp.Server/Program.cs) to announce the tools and instruct the LLM to use them.
|
||||
You can then register the MCP server in your tool of choice. While the exact steps may vary depending on your IDE,
|
||||
here are some common configurations.
|
||||
|
||||
While the MCP prompt is elaborate, you may need to be explicit in prompts and for example, add "Use Duende samples" when looking for code samples.
|
||||
### Visual Studio / Visual Studio Code
|
||||
|
||||
## Run and Register the MCP
|
||||
You can register the Duende Documentation MCP Server
|
||||
[in your user settings](https://code.visualstudio.com/docs/copilot/chat/mcp-servers#_add-an-mcp-server-to-your-user-settings)
|
||||
to make it available in any workspace.
|
||||
|
||||
### Visual Studio Code
|
||||
|
||||
Add a `.vscode/mcp.json` to your workspace:
|
||||
Alternatively, you can add a `.vscode/mcp.json` file to your workspace:
|
||||
|
||||
```json
|
||||
{
|
||||
"servers": {
|
||||
"duende-mcp-hosted": {
|
||||
"duende-mcp": {
|
||||
"type": "stdio",
|
||||
"command": "dnx",
|
||||
"args": ["Duende.Documentation.Mcp.Server@1.0.0", "--yes"],
|
||||
"args": ["Duende.Documentation.Mcp.Server", "--yes"],
|
||||
"env": {}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
Open GitHub Copilot and select Agent Mode to work with the MCP server.
|
||||
|
||||
### JetBrains Rider
|
||||
|
||||
In JetBrains Rider settings, navigate to **Tools \| AI Assistant \| Model Context Protocol (MCP)**.
|
||||
Next, add a new MCP server. In the dialog that opens, select **As JSON** and enter the following configuration:
|
||||
|
||||
```json
|
||||
{
|
||||
"mcpServers": {
|
||||
"duende-mcp": {
|
||||
"command": "dnx",
|
||||
"args": ["Duende.Documentation.Mcp.Server", "--yes"]
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
Set the working directory to a path on your machine where the Duende Documentation MCP Server can store its database
|
||||
index. Not setting the working directory will result in the MCP server failing to start because it cannot create the
|
||||
database file.
|
||||
|
||||
## Tools and Example Prompts
|
||||
|
||||
The Duende Documentation MCP Server has several tools available:
|
||||
|
||||
* Free text search on blogs, docs, or samples
|
||||
* Fetch specific page
|
||||
* Get all content for a sample
|
||||
* Get a specific file from a sample
|
||||
|
||||
The Duende Documentation MCP Server has [instructions](src/Duende.Documentation.Mcp.Server/Program.cs) to announce the
|
||||
tools it provides, and instructs the LLM to use them. While this MCP prompt is elaborate, you may need to be explicit
|
||||
in prompts and for example, add "Use Duende samples" when you expect to update code with your LLM.
|
||||
|
||||
Example prompts:
|
||||
|
||||
* "What is a client in OpenID Connect?"
|
||||
* "What is automatic key management?"
|
||||
* "How can I validate a JWT token in ASP.NET Core?"
|
||||
* "What is a Personal Access Token and how do I create one?"
|
||||
|
||||
Sometimes, it may be necessary to provide more context to the LLM. For example, when you want to know more about a
|
||||
specific topic you expect in the Duende documentation or blog, you can instruct the LLM to use the Duende Documentation
|
||||
MCP Server:
|
||||
|
||||
* "Explain .NET TLS certificates - use Duende"
|
||||
* "Can I add passkeys to Razor Pages? use Duende"
|
||||
|
||||
## Support
|
||||
|
||||
If you experience an issue with the Duende Documentation MCP or have any other feedback, please open an issue in our
|
||||
[Duende community](https://duende.link/community).
|
||||
|
||||
## Technical details
|
||||
|
||||
### Development
|
||||
|
||||
* Run the project. This will host a server on port 3000 (http), and with stdio bindings.
|
||||
|
|
@ -48,23 +105,29 @@ Add a `.vscode/mcp.json` to your workspace:
|
|||
}
|
||||
```
|
||||
|
||||
## Indexers
|
||||
### Indexers
|
||||
|
||||
The project uses full-text search with SQLite. There are indexes for docs, blog, and samples. Indexes are built by dedicated background services.
|
||||
The project uses full-text search with SQLite. There are indexes for docs, blog, and samples. Indexes are built by
|
||||
dedicated background services.
|
||||
|
||||
### Docs
|
||||
#### Docs
|
||||
|
||||
Documentation is indexed by parsing LLMs.txt hosted on Duende documentation: https://docs.duendesoftware.com/llms.txt
|
||||
|
||||
LLMs.txt is a technique that allows LLMs to find information in a Markdown-based format, so it can be parsed more easily (and within the LLM context).
|
||||
While available on many websites, none of the vendors currently support this out-of-the-box.
|
||||
LLMs.txt is a technique that allows LLMs to find information in a Markdown-based format, so it can be parsed more
|
||||
easily (and within the LLM context). While available on many websites, none of the vendors currently support this
|
||||
out-of-the-box.
|
||||
|
||||
### Blog
|
||||
#### Blog
|
||||
|
||||
Blogs are indexed by parsing the RSS feed of the at https://duendesoftware.com/blog/, and then fetching each page and converting it into markdown.
|
||||
Blogs are indexed by parsing the RSS feed of the at https://duendesoftware.com/blog/, and then fetching each page and
|
||||
converting it into markdown.
|
||||
|
||||
### Samples
|
||||
#### Samples
|
||||
|
||||
Samples are indexed by looking at the samples-specific LLMs.txt file at https://docs.duendesoftware.com/_llms-txt/identityserver-sample-code.txt
|
||||
Samples are indexed by looking at the samples-specific LLMs.txt file
|
||||
at https://docs.duendesoftware.com/_llms-txt/identityserver-sample-code.txt
|
||||
|
||||
This document contains sample names and descriptions, and includes links to GitHub. The GitHub repository is downloaded as an archive (https://github.com/duendesoftware/samples/archive/refs/heads/main.zip), and as part of indexing all of the `.cs`, `.cshtml` and relevant `.js` files are added to the index. Other files are ignored.
|
||||
This document contains sample names and descriptions, and includes links to GitHub. The GitHub repository is downloaded
|
||||
[as an archive](https://github.com/duendesoftware/samples/archive/refs/heads/main.zip), and all `.cs`, `.cshtml`
|
||||
and relevant `.js` files are added to the index. Other files are ignored.
|
||||
|
|
|
|||
Loading…
Reference in a new issue