@sensegrep/mcp

@sensegrep/mcp

MCP server exposing sensegrep capabilities to AI coding assistants.

Install

npm install -g @sensegrep/mcp

MCP Configuration

{
  "mcpServers": {
    "sensegrep": {
      "command": "npx",
      "args": ["-y", "@sensegrep/mcp"]
    }
  }
}

Or with global install first:

npm install -g @sensegrep/mcp

{
  "mcpServers": {
    "sensegrep": {
      "command": "sensegrep-mcp"
    }
  }
}

Tools

Canonical tool names:

• sensegrep_search
• sensegrep_show
• sensegrep_literal
• sensegrep_context
• sensegrep_survey
• sensegrep_cluster
• sensegrep_graph
• sensegrep_index
• sensegrep_detect_duplicates

Search returns schema v2 compact cards by default. Call sensegrep_show with a selected card id, or set resultDetail to content/full, to expand only the evidence an agent needs. MCP publishes an output schema and returns the complete payload once in structuredContent; textual content is only a short transport summary. Results report retrieval sufficiency, compact index state, applicable budgets, and structured warnings.

Legacy aliases are still accepted for compatibility:

• sensegrep.search
• sensegrep.index
• sensegrep.detect_duplicates

Environment Variables

Common MCP/runtime variables:

• SENSEGREP_ROOT - root directory to index/search
• SENSEGREP_WATCH - watcher toggle (0, false, off, no disables)
• SENSEGREP_PROVIDER - ollama, gemini, openai, or bedrock
• SENSEGREP_EMBED_MODEL, SENSEGREP_EMBED_DIM
• GEMINI_API_KEY / GOOGLE_API_KEY
• SENSEGREP_OPENAI_API_KEY / FIREWORKS_API_KEY / OPENAI_API_KEY

For the complete and official list, see docs/mcp-setup.md in the repository.

Documentation

• MCP setup: https://github.com/Stahldavid/sensegrep/blob/main/docs/mcp-setup.md
• Repository: https://github.com/Stahldavid/sensegrep