yigitkonur/mcp-latitude-prompts

MCP server that bridges your AI assistant to Latitude.so prompt management. list, read, run, push, and pull PromptL prompts — all from inside a conversation. like prompt DevOps without leaving your editor.

npx mcp-latitude-prompts

what it does

7 tools exposed over MCP stdio transport. your AI assistant gets full access to your Latitude prompt library.

• pre-validation — every prompt is parsed with promptl-ai before push. if anything has errors, nothing gets deployed
• 3-step transactional deploy — draft, push with SHA-256 content hashes, publish to LIVE
• diff engine — only changed prompts get pushed, compared by content hash
• dynamic tool descriptions — run_prompt and add_prompt descriptions are built at startup with your actual prompt names and their {{ variables }}
• 30-second prompt cache — force-refreshed after any mutation
• 52-topic docs system — PromptL syntax, techniques (ReAct, CoT, ToT, RAG, etc.), recipes, provider configs

install

npm install -g mcp-latitude-prompts

or just run it directly:

npx mcp-latitude-prompts

or build from source:

git clone https://github.com/yigitkonur/mcp-latitude-prompts.git
cd mcp-latitude-prompts
pnpm install && pnpm build

configure

two required env vars:

export LATITUDE_API_KEY="your-api-key"       # from https://app.latitude.so/settings/api-keys
export LATITUDE_PROJECT_ID="your-project-id"

config is loaded in priority order: ~/.mcp/configs.json < .env file < direct env vars.

global config file

optional. place at ~/.mcp/configs.json:

{
  "latitude": {
    "environments": {
      "LATITUDE_API_KEY": "your-key",
      "LATITUDE_PROJECT_ID": "your-project-id"
    }
  }
}

usage with Claude Desktop

add to your MCP client config:

{
  "mcpServers": {
    "latitude": {
      "command": "npx",
      "args": ["-y", "mcp-latitude-prompts"],
      "env": {
        "LATITUDE_API_KEY": "your-api-key",
        "LATITUDE_PROJECT_ID": "your-project-id"
      }
    }
  }
}

development

pnpm install
pnpm build                # compile TypeScript
pnpm mcp:stdio            # build + run stdio
pnpm mcp:inspect          # build + run with MCP inspector at localhost:3000
pnpm dev:stdio            # run with @modelcontextprotocol/inspector

logs go to ~/.mcp/data/<package-name>.<session-uuid>.log — stdout is reserved for the MCP stdio stream.

how push works

the deploy pipeline is transactional:

1. read all .promptl files from your folder
2. validate every prompt with promptl-ai — abort on any error
3. compute diff against current LIVE (SHA-256 content hashes)
4. POST /versions — create a named draft
5. POST /versions/:uuid/push — submit changes
6. POST /versions/:commitUuid/publish — publish to LIVE

add_prompt uses the same pipeline but merges new prompts into a temp copy of the existing set first — so nothing gets deleted.

project structure

src/
  index.ts        — entry point, env validation
  server.ts       — MCP server factory + stdio transport
  tools.ts        — all 7 tool definitions and handlers
  api.ts          — Latitude HTTP client, diff engine, validation
  types.ts        — TypeScript interfaces
  docs/
    index.ts      — 52-topic docs map, search, and lookup
    types.ts      — doc topic types
    help.ts       — server overview content
    metadata.ts   — topic search metadata
    core-syntax.ts, phase1.ts, phase2.ts, phase3.ts, techniques.ts, recipes.ts

troubleshooting

license

MIT