AI-Assisted Development
ErisPulse provides two complementary ways for AI-assisted development, enabling AI to generate code based on the latest framework specifications:
- Material Documentation: A large Markdown file, loaded as context in one go, suitable for full project development.
- MCP Server: Allows AI to retrieve official documentation on-demand, suitable for daily code completion and API lookup.
| Material Documentation | MCP Server | |
|---|---|---|
| Format | A large Markdown file, loaded in one go | AI retrieves on-demand |
| Context Cost | High (consumes tokens) | Low (only fetches relevant snippets) |
| Timeliness | Updated with version releases | Real-time (GitHub pull + cache) |
| Suitable For | Large context window, full project development | Daily code completion, API lookup |
| Client | Any AI tool | Only clients supporting MCP |
The two approaches are not mutually exclusive: for large projects, both can be used together—material documentation for the base, and MCP as a safety net for missing information.
Material Documentation
The material documentation is located in the prompts/ directory and is divided into three types based on development scenarios:
| Document | Scenario | Description |
|---|---|---|
| ErisPulse-ModuleDev.md | Module Development | Covers the entire module development workflow (event handling, routing, lifecycle, etc.) |
| ErisPulse-AdapterDev.md | Adapter Development | Builds on module development and adds core adapter concepts, SendDSL, and platform adaptation guidelines |
| ErisPulse-Full.md | Full Stack Reference | All the above content plus a complete user guide and API reference collection |
Access method: Retrieve from the prompts/ directory in the docs-generated branch (updated synchronously with the documentation), or download via the "AI-Assisted Development" page on the official website.
Usage
- Choose the appropriate document based on your target (module →
ModuleDev, adapter →AdapterDev, complex needs →Full) - Provide the document content as context to the AI: place it in your IDE workspace, paste directly into chat-style interfaces, or inject as a system message in API calls
- Describe your requirements using the template below; the more specific your description, the higher the quality of the generated output
Requirement Description Template
Module:
Please generate a complete code for a [Module Name] module based on the ErisPulse module development specification.
Function Description: [Core Functionality]
Events to listen for: [Message / Command / Notification / Request], Processing Logic: [Operation]
Required Configuration Items: [Key Name]: [Purpose] ([Required/Optional], Default Value: [Value])
Other Requirements: [Additional Constraints]
Adapter:
Please generate a complete code for an [Adapter Name] adapter based on the ErisPulse adapter development specification.
Platform Information: [Name], Communication Protocol: [WebSocket / WebHook / HTTP Polling], API Documentation: [URL]
Event Mapping: Platform Event [Type] → OneBot12 Mapping [Relationship]
Send Methods to Implement: [Text / Image / Voice ...]
Configuration Items: [Key Name]: [Purpose] ([Required/Optional])
Complete Documentation Package
The skills.zip file packages all the original documentation in this language (unmerged and unmodified), with a README index at the root directory: documents grouped by directory, each entry marked with "When to read this document." An AI can load only the required documents according to the index.
Difference from Material Documentation
| Material Documentation | Complete Documentation Package | |
|---|---|---|
| Format | Single .md merged document |
All original documents + index README |
| Usage | Entire document inserted into context | Unzip and place into workspace; AI reads documents as needed |
| Applicable | When context window is sufficient, for whole project development | Long-term workspace residency, document retrieval on demand |
Both are generated from the same batch of documents and are updated synchronously.
Download
Download from the "AI-assisted Development" page on the official website, or directly access the docs/en/ai-support/skills.zip file from the docs-generated branch.
MCP Server
ErisPulse provides a documentation retrieval MCP Server, allowing AI coding assistants (any that support MCP) to directly search and access ErisPulse's official documentation while writing code.
There are two ways to connect: local process (recommended) or official online endpoint. The toolset is identical:
| Tool | Parameters | Description |
|---|---|---|
search_docs |
query (required), top_k?=5, lang?=zh-CN |
BM25 keyword search, multiple keywords can be passed at once |
read_document |
doc_path (required), lang?=zh-CN |
Reads the full Markdown of a single document |
list_documents |
lang?=zh-CN |
Lists all document titles, paths, and categories in the current language |
list_languages |
— | Lists all languages supported by the documentation and their document counts |
Supported languages: zh-CN / en / zh-TW / ja / ru. Search tips: use multiple keywords rather than full sentences, for example, command registration event listening is better than how to register a command.
Method 1: Local Process (Recommended)
Install and start locally:
npm i -g @erisdev/mcp-server # Global installation → `epsdk-mcp` command appears
epsdk-mcp --server # Start the MCP service; first run automatically pulls documentation to ~/.cache/erispulse-mcp/docs
Alternatively, run without global installation using npx:
npx @erisdev/mcp-server --server
Specify a local documentation directory (offline mode):
ERISPULSE_DOCS_DIR=/path/to/erispulse/docs epsdk-mcp --server
Configure in the MCP client using stdio transmission with standard key-value pairs (command + args pointing to the above startup method; specific configuration locations and formats are detailed in each client's MCP integration documentation):
{
"command": "epsdk-mcp",
"args": ["--server"]
}
or
{
"command": "npx",
"args": ["@erisdev/mcp-server", "--server"]
}
Method 2: Official Online Endpoint
If you don't want to start a local process, connect to the official hosted mcp.erisdev.com. Configure the MCP client using HTTP transmission:
{
"url": "https://mcp.erisdev.com/"
}
The service is publicly accessible by default, without a token, with IP rate limiting (60 requests per minute per IP). The official endpoint can be disabled at any time; for production workflows, it is recommended to use Method 1 (local process) or self-hosting.
Frequently Asked Questions
The generated code does not meet expectations? Check if the full documentation is provided; add more details in the requirements (input/output examples, boundary conditions); have the AI generate step-by-step (first skeleton, then features); refer to the examples/ directory for example contexts.
After MCP integration, the AI did not call search_docs?
Confirm that the client has loaded the server (the tools will be effective after restarting the client); some clients require explicit prompts in the prompt, such as "use the ErisPulse documentation tool to verify the API."
Next Steps
- Module Development Guide -- Complete tutorial for manual module development
- Adapter Development Guide -- Complete tutorial for manual adapter development
- Example Code -- Reference existing module and adapter implementations
- Module Builder -- A browser-based visual AI module generator