Pampa

Provides an MCP server to index, search, and retrieve code with semantic memory for AI agents.

javascript

25

GitHub Stars

javascript

Language

4 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "tecnomanu-pampa": {
      "command": "npx",
      "args": [
        "-y",
        "pampa",
        "mcp"
      ]
    }
  }
}

You can run PAMPA as an MCP server to enable AI agents to search, retrieve, and learn from your codebase’s memory. By exposing semantic search and incremental indexing through a lightweight local setup, you gain fast, accurate code understanding without exposing or re-scanning large trees on every query.

How to use

You interact with PAMPA through an MCP client to search code semantically, fetch full code chunks by SHA, and manage indexing. Start by ensuring your project is indexed, then let PAMPA keep memory up to date as you evolve the codebase. You can search with natural language like “authentication function” and receive ranked, context-aware results. Use context packs to narrow searches to specific modules or languages, and leverage the cross-encoder reranker for higher precision when needed.

How to install

Prerequisites: you need Node.js installed (version 20+ recommended). You also use MCP clients that speak the standard MCP protocol. The server runs locally via an MCP-compatible interface and does not require a separate runtime beyond Node.js.

Step by step you can set up PAMPA and its MCP server integration:

Install the CLI and run the MCP server

# Install locally and expose the MCP server via npx
npx pampa mcp

Configure your MCP client to talk to PAMPA

You configure your MCP client to invoke the PAMPA MCP server using the standard npx-based command, ensuring the client can start a local server process as needed.

Starting and using the MCP server directly

To start the PAMPA MCP server directly from the command line, use the following standard pattern that the client expects. This starts a stdio server so your MCP client can communicate with PAMPA without a remote endpoint.

npx pampa mcp

Available tools

search_code

Semantically searches the indexed codebase and returns matching code chunks with similarity scores. Supports optional provider overrides, scoped filters, and ranking controls.

get_code_chunk

Retrieves the complete source code for a specific chunk using its SHA.

index_project

Indexes a project, creating the local embeddings database and chunk storage, and generating a codemap for version-tracking.

update_project

Updates the index after code changes to keep AI memory synchronized with the current repository state.

get_project_stats

Provides statistics about the indexed project, such as language distribution and file counts.