120 lines
2.4 KiB
Markdown
120 lines
2.4 KiB
Markdown
# Meta MCP (ContextVM)
|
|
|
|
Meta MCP is a ContextVM server that accepts **requirements** for new MCP tools and returns a job id. It is intended to be used by other AIs that only have access to a single "meta" tool. The meta MCP then builds, deploys, and announces a new MCP server via Nostr (ContextVM).
|
|
|
|
This repo contains the **server-side** MCP for the meta tool. It implements the request + job tracking surface and a simple build/deploy pipeline (scaffold → build → announce via CVMI).
|
|
|
|
---
|
|
|
|
## Goals
|
|
|
|
- Accept requirements for new tools
|
|
- Create a repo/code for the new MCP server
|
|
- Build + run the MCP server on this VPS
|
|
- Announce the server on Nostr (ContextVM)
|
|
- Return the **server pubkey** back to the requester
|
|
|
|
---
|
|
|
|
## Tool API
|
|
|
|
### `meta.request_tool`
|
|
Request a new MCP tool/server.
|
|
|
|
**Input**
|
|
```json
|
|
{
|
|
"requirements": [
|
|
{
|
|
"name": "get_current_weather_by_city",
|
|
"description": "Returns current weather for a city",
|
|
"inputSchema": {"type":"object","properties":{"city":{"type":"string"}},"required":["city"]}
|
|
}
|
|
],
|
|
"timeoutSeconds": 120,
|
|
"requester": {"name":"weather-bot","pubkey":"npub1..."},
|
|
"context": {"notes":"Need fast response"}
|
|
}
|
|
```
|
|
|
|
**Output**
|
|
```json
|
|
{
|
|
"status": "building",
|
|
"message": "Accepted. Building MCP server (pipeline stub).",
|
|
"jobId": "<uuid>"
|
|
}
|
|
```
|
|
|
|
### `meta.job_status`
|
|
Check a previous job.
|
|
|
|
**Input**
|
|
```json
|
|
{ "jobId": "<uuid>" }
|
|
```
|
|
|
|
### `meta.jobs_recent`
|
|
List recent jobs.
|
|
|
|
**Input**
|
|
```json
|
|
{ "limit": 10 }
|
|
```
|
|
|
|
---
|
|
|
|
## Development
|
|
|
|
```bash
|
|
npm install
|
|
npm run dev
|
|
```
|
|
|
|
---
|
|
|
|
## Deployment (ContextVM/CVMI)
|
|
|
|
This server runs via stdio. Use **cvmi serve** to expose it over Nostr.
|
|
|
|
Example:
|
|
```bash
|
|
cvmi serve -- node dist/server.js
|
|
```
|
|
|
|
You can set `CVMI_SERVE_*` env vars to configure relays, encryption, and keys.
|
|
|
|
---
|
|
|
|
## Pipeline
|
|
|
|
The pipeline is currently minimal and runs locally:
|
|
|
|
1) Scaffold a new MCP server in `outputDir`
|
|
2) `npm install` + `npm run build`
|
|
3) Run `npx cvmi serve -- node dist/server.js`
|
|
4) Parse and store the announced pubkey
|
|
|
|
Configuration is read from `/root/.meta-mcp/config.json` (or `META_MCP_CONFIG`).
|
|
|
|
Example config:
|
|
```json
|
|
{
|
|
"outputDir": "/root/.openclaw/workspace/generated",
|
|
"cvmi": { "seedPath": "/root/.cvmi/seed.json" }
|
|
}
|
|
```
|
|
|
|
## Next Steps
|
|
|
|
- Add persistent storage for job state
|
|
- Repo creation + git push for generated MCPs
|
|
- Systemd services per generated MCP
|
|
- Security controls / rate limits
|
|
|
|
---
|
|
|
|
## License
|
|
|
|
MIT
|