Tugra MCP server
{
"mcpServers": {
"tugra-ai": {
"command": "npx",
"args": [
"-y",
"tugra@latest"
]
}
}
}{
"mcpServers": {
"tugra-ai": {
"command": "npx",
"args": [
"-y",
"tugra@latest"
]
}
}
}{
"mcpServers": {
"tugra-ai": {
"command": "npx",
"args": [
"-y",
"tugra@latest"
]
}
}
}{
"servers": {
"tugra-ai": {
"command": "npx",
"args": [
"-y",
"tugra@latest"
]
}
}
}{
"mcpServers": {
"tugra-ai": {
"command": "npx",
"args": [
"-y",
"tugra@latest"
]
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall npx -y tugraversion 0.1.4
What it does
Tugra is a local, file-based memory system for AI agents where every stored claim includes its source, verification date, shelf life, and optional boundaries against invention. It runs as an MCP server with no cloud sync or automatic merging, designed for developers building agents that need to track fact provenance.
- Search a vault of facts ranked by relevance, freshness, and confidence, with retired or expired facts excluded by default.
- Store claims with attached metadata about where they came from and when they were last verified, then read individual facts by ID.
- Propose new facts while rejecting attempts to write secrets or boundary-marked content that should not be invented by the model.
Tools this server exposes
| Tool | Does |
|---|---|
| fact_search | Search the vault for facts, ranked by token score, freshness, and confidence; optionally include archived or expired facts. |
| fact_read | Retrieve a single fact by its unique ID with content escaped before delivery to the model. |
| fact_propose | Write a draft fact to the vault, rejecting attempts to store secret patterns or boundary-marked content. |
| event_report | Append a local telemetry line to the events directory without sending data over the network. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.