What can you do with this server?

This MCP server indexes project documentation, source code, and structured logs into a local vector database and exposes tools to search, inspect, compare, and manage indexed projects. * Semantic search across documentation, specs, and code, with project and file-type filters. * Exact text lookup for hex patterns, variables/constants, and function definitions/usages. * Retrieve full documents, list projects/files, get project summaries, and compare the same concept across two projects. * Manage indexing by adding projects, files, folders, and glob patterns; remove projects, clear project indexes, or remove individual files from the index. * Search indexed logs with semantic query plus filters for severity, time range, error-code prefix, device ID, and event type. * Index specific log files or time windows on demand.

How do I use pow-mcp-rag-new?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@pow-mcp-rag-new find the function that indexes PDFs" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

pow-mcp-rag-new

by powmarx

Overview Schema Source Related Servers Discussions Admin

Python

Local

pow-mcp-rag-new

A local RAG (Retrieval-Augmented Generation) MCP server for project documentation and code.

Overview

This project provides a Model Context Protocol (MCP) server that indexes project documentation (specs, headers, source files, PDFs, configs) into a local vector database (ChromaDB) and exposes semantic search tools to Kiro or any MCP-compatible client.

Supports multiple tech stacks: C/C++, Python, Go, C#, Node.js/TypeScript.

Distribution name: pow-rag-mcp (unrelated to rag-mcp or rag-mcp-server packages).

Minimum Python version: 3.11 (matches requires-python = ">=3.11" in pyproject.toml).

Three deployment modes:

Docker (Phase 1) — zero local Python needed; server runs in a container
PyPI (recommended for new users) (Phase 2a) — install via uvx --from, uv tool install, or pip install from PyPI; no repo checkout or local index needed
Local PyPI + uvx (Phase 2b) — install rag-mcp via uvx from a local package index; no persistent venv, easiest to keep updated. Stepping stone toward a hosted index.
pip install (Phase 2c) — install rag-mcp directly into a Python environment.
llamaindex (Phase 3): Using of llamaindex to handle several different data sources.

Related MCP server: ragi

License

This project is licensed under the Apache 2.0 License — see the LICENSE file for details.

Quick Start

PyPI (recommended for new users)

Install from PyPI using your preferred method:

Option 1: One-off execution with uvx

uvx --from pow-rag-mcp rag-mcp serve

Option 2: Persistent install with uv tool install

uv tool install pow-rag-mcp

Option 3: Traditional pip

pip install pow-rag-mcp

All three methods fetch pow-rag-mcp from PyPI directly — no repository checkout or local package index is required.

See doc/PIP_INSTALL_GUIDE.md for full details (config seeding, bundled docs, upgrades, and migrating to a hosted index).

Local PyPI + uvx (recommended for development/publishing workflows)

cd <your-checkout>/pow-mcp-rag-new
setup-pypi.bat

This builds the wheel, publishes it to a local pypiserver index (packages/), installs it as a persistent uv tool (a stable exe at ~/.local/bin/rag-mcp.exe — resolved once, not on every launch), and wires up ~/.kiro/settings/mcp.json + .vscode/mcp.json to launch it directly. No Docker, no persistent venv to manage. See TROUBLESHOOTING.md for why this is preferred over plain uvx --from on Windows with this package's large dependency tree.

Full guide: doc/PIP_INSTALL_GUIDE.md

Docker (alternative)

cd <your-checkout>/pow-mcp-rag-new
docker build -t rag-mcp-new-pip:latest .

# Index your projects (set SRC to your repos folder)
$SRC = "C:/Users/you/GIT"   # PowerShell
docker run --rm -v "${SRC}:/projects:ro" -v rag-mcp-new-pip-data:/app/data rag-mcp-new-pip:latest python indexer.py

Then add to ~/.kiro/settings/mcp.json:

{
  "mcpServers": {
    "rag-mcp": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-v", "C:/Users/you/GIT:/projects:ro",
        "-v", "rag-mcp-new-pip-data:/app/data",
        "rag-mcp-new-pip:latest",
        "python", "server.py", "--no-reindex"
      ],
      "disabled": false,
      "autoApprove": [
        "search_docs", "search_specs", "search_code", "search_logs",
        "list_projects", "list_files", "get_document", "get_project_summary",
        "find_function", "find_variable", "search_hex_pattern", "compare_projects",
        "add_project", "add_file", "add_folder", "add_pattern", "index_log_file"
      ]
    }
  }
}

Restart Kiro — the RAG is ready. Full guide: doc/DOCKER_GUIDE.md (project management, HTTP server, docker run CLI reference, setup-docker.bat automation).

pip install (hosted index, once available)

pip install torch --index-url https://download.pytorch.org/whl/cpu
rag-mcp config   # seeds config on first run, shows resolved paths
rag-mcp index
rag-mcp serve

Full guide: doc/PIP_INSTALL_GUIDE.md

Documentation

Reading this from pip show / a package-index page instead of the repo? This README's links are relative paths into the pow-mcp-rag-new repo (GitHub/clone) and won't resolve from an installed package alone. Five docs travel with the install and are available offline via rag-mcp docs <name> (see table below); the rest require the repo checkout.

Guide	Covers	Bundled in package?
doc/DOCKER_GUIDE.md	Full Docker setup, adding/removing projects, HTTP server, complete CLI reference (Docker mode)	No — repo only
doc/PIP_INSTALL_GUIDE.md	Local PyPI + uvx setup, pip installation, config seeding, building/publishing the wheel	No — repo only
doc/ARCHITECTURE.md	How indexing/retrieval works, embedding model, PDF handling, file structure	No — repo only
doc/CLI_REFERENCE.md	Full `rag-mcp` CLI reference: `serve`/`index`/`config`/`docs`/`get_metrics`, all flags and env vars	Yes — `rag-mcp docs cli`
doc/TOOLS_GUIDE.md	Full MCP tool reference (21 tools) with usage examples	Yes — `rag-mcp docs tools`
doc/LOG_INDEXING_GUIDE.md	Structured log indexing usage guide	Yes — `rag-mcp docs log-indexing`
doc/LOG_PATTERN_CONFIGURATION.md	How to write pattern configs for new log formats	Yes — `rag-mcp docs log-patterns`
doc/OBSERVABILITY.md	Structured JSON event log (tool calls, embedding failures, backfills), counters, `get_metrics` tool and CLI	Yes — `rag-mcp docs observability`
doc/TROUBLESHOOTING.md	Common issues and fixes	No — repo only

Run rag-mcp docs with no arguments to list the bundled docs from any install (pip, uvx, or uv tool install) without needing the repo.

What gets indexed

File types are configured in config.yaml under index_extensions. By default: C/C++, Python, React/JS/TS, C#, Go, Kotlin, Markdown, PDF, text. Directories in excluded_dirs (build, node_modules, .git, ...) are skipped. See doc/DOCKER_GUIDE.md for details on configuring projects and adding new sources.

MCP Tools

Once configured, 21 MCP tools are available for searching, browsing, and managing your indexed projects — see doc/TOOLS_GUIDE.md for the full list with examples (also available offline: rag-mcp docs tools).

Key tools:

search_docs / search_specs / search_code — semantic search, scoped by file type
search_hex_pattern, find_function, find_variable — exact text matching for codes/symbols
search_logs, index_log_file — structured log search and on-demand indexing
add_project, add_pattern, remove_project — index management from Kiro chat
get_metrics — live counters (tool calls, embedding failures, backfills); see doc/OBSERVABILITY.md

Log Indexing

Structured log indexing with severity filtering, time-window search, and hex error-code matching. The pipeline is fully generic — driven by YAML pattern configs, so any log format can be supported without code changes. See doc/DOCKER_GUIDE.md and doc/LOG_INDEXING_GUIDE.md (also: rag-mcp docs log-indexing).

Need Help?

See doc/TROUBLESHOOTING.md for common issues, including:

MCP server hangs on startup
Stale search results after re-indexing
entrypoint.sh exec errors from CRLF line endings
--project <NAME> silently doing nothing
Concurrent query errors

Available Tools

20 tools

add_fileA

Index a specific file into an existing project and persist it in config.yaml so it gets re-indexed automatically on future runs.

Args: file_path: Absolute path to the file to index project: Project name to add the file to (must already exist)

Returns: Confirmation with chunk count, or error message

Inputs: project (required), file_path (required). Returns: result

add_folderA

Index all files in a folder into an existing project and persist the pattern in config.yaml so the folder gets re-indexed automatically on future runs.

Args: folder_path: Absolute path to the folder to index project: Project name to add the folder to (must already exist) pattern: Glob pattern for files within the folder (default: "**/*" for all files)

Returns: Confirmation with file/chunk counts, or error message

Inputs: project (required), folder_path (required), pattern. Returns: result

add_patternA

Add a glob pattern to an existing project and index matching files immediately. Persists the pattern in config.yaml so it gets re-indexed automatically on future runs.

Use this when you want to index a specific set of files by pattern rather than pointing at a concrete folder path. The pattern is relative to the project base_path.

Args: project: Project name to add the pattern to (must already exist) pattern: Glob pattern relative to project base_path (e.g., "doc/specifications/**/*.md") type: File type classification — one of: source, header, documentation, config (default: documentation) description: Optional human-readable description for this source entry

Returns: Confirmation with file/chunk counts, or error message

Examples: add_pattern("my-project", "doc/specifications//*.md", "documentation", "Hardware specs") add_pattern("my-project", "src//*.json", "config", "JSON config files")

Inputs: pattern (required), project (required), type, description. Returns: result

add_projectA

Add a new project to the RAG index. Auto-detects source patterns and indexes immediately.

Args: name: Unique project name (e.g., "my-new-api") path: Absolute path to the project root directory

Returns: Summary of detected patterns and indexing result

Inputs: name (required), path (required). Returns: result

clear_project_indexA

Clear all indexed data for a project without removing it from config.yaml. The project remains configured and can be re-indexed later.

Args: name: Project name to clear (e.g., "my_device_logs")

Returns: Confirmation with number of chunks deleted.

Inputs: name (required). Returns: result

compare_projectsA

Search for the same concept in two projects side by side. Useful for comparing implementations across two versions or variants of the same project.

Args: query: What to search for (e.g., "dispense command flow") project_a: First project name project_b: Second project name top_k: Results per project (default: 3)

Returns: Side-by-side results from both projects

Inputs: query (required), project_a (required), project_b (required), top_k. Returns: result

find_functionA

Find where a function is defined (headers) and used (source files).

Args: function_name: Name of the function to find (e.g., "StoreMoney", "CmdDispense") project: Optional project name filter

Returns: Header declarations and source usages of the function

Inputs: function_name (required), project. Returns: result

find_variableB

Find where a specific variable, constant, enum value, or #define is defined and used. Uses text matching to find exact occurrences across indexed files.

Args: variable_name: Name of the variable/constant to find project: Optional project name filter

Returns: Files containing the variable with context snippets

Inputs: variable_name (required), project. Returns: result

get_documentA

Retrieve all indexed chunks of a specific document.

Args: file_path: Relative file path within the project (e.g., "src/Device.h") project: Project name (e.g., "my-project")

Returns: Full reconstructed document content from indexed chunks

Inputs: project (required), file_path (required). Returns: result

get_project_summaryA

Get a quick overview of an indexed project: description, file counts by type, total chunks.

Args: project: Project name

Returns: Project summary with statistics

Inputs: project (required). Returns: result

index_log_fileA

Index a specific log file (or time window within it) on demand. Use this instead of waiting for background reindex on large log files.

Args: file: Log filename or glob pattern (e.g., "device-26-04-28.log", "device-2026-05-14*.log") project: Project name containing log sources (required) time_from: Optional start time filter, HH:MM:SS format (e.g., "14:00:00"). Only index events at or after this time. time_to: Optional end time filter, HH:MM:SS format (e.g., "15:30:00"). Only index events at or before this time.

Returns: Summary of indexing results (files processed, chunks created, time range)

Inputs: file (required), project, time_to, time_from. Returns: result

list_filesA

List all indexed files for a project, optionally filtered by type.

Args: project: Project name (e.g., "my-project") file_type: Optional filter: "header", "source", "documentation", "config"

Returns: List of indexed file paths with their types

Inputs: project (required), file_type. Returns: result

list_projectsA

List all indexed projects and their statistics.

Returns: List of projects with document counts and descriptions

Inputs: no named parameters. Returns: result

remove_file_from_indexA

Remove all indexed chunks for a specific file from a project's index. Useful for removing large log files or outdated files without clearing the entire project.

Args: file_path: Relative file path within the project (e.g., "device-26-04-28.log") project: Project name (e.g., "my_device_logs")

Returns: Confirmation with number of chunks removed.

Inputs: project (required), file_path (required). Returns: result

remove_projectA

Remove a project from the RAG index. Deletes all indexed chunks and marks the project as removed in config.yaml (preserving configuration for future re-add).

Args: name: Project name to remove (e.g., "my_device_logs")

Returns: Confirmation of removal with chunk count deleted.

Inputs: name (required). Returns: result

search_codeA

Search only source code and header files (filters out documentation). Use when you need implementation details.

Args: query: Natural language search query project: Optional project name filter top_k: Number of results (default: 5, max: 20) headers_only: If true, search only header files (API contracts)

Returns: Relevant source code chunks

Inputs: query (required), top_k, project, headers_only. Returns: result

search_docsA

Semantic search across indexed project documentation.

Args: query: Natural language search query (e.g., "how does dispense rejection work") project: Optional project name to filter results. Leave empty to search all projects. top_k: Number of results to return (default: 5, max: 20) file_type: Optional filter by file type: "header", "source", "documentation", "config"

Returns: Relevant document chunks with metadata (file path, project, type)

Inputs: query (required), top_k, project, file_type. Returns: result

search_hex_patternA

Search for hex error codes or patterns in indexed documents using text matching. Supports partial matches (e.g., '88a153' finds 88a15300, 88a15310, etc.) Use this for device error codes, packet IDs, or any hex value lookup.

Args: pattern: Hex pattern to search for (e.g., "88a153", "0x0521", "ERROR_118") project: Optional project name to filter results. Leave empty to search all. top_k: Maximum results to return (default: 10)

Returns: Document chunks containing the hex pattern with file and project info

Inputs: pattern (required), top_k, project. Returns: result

search_logsA

Search indexed log events with structured filtering and semantic search.

Args: query: Semantic search query (max 512 chars). Optional if filters provided. project: Filter by project name. severity: Filter by severity (DEBUG, INFO, WARNING, ERROR, CRITICAL). time_range_start: ISO 8601 start time filter. time_range_end: ISO 8601 end time filter. error_code_pattern: Prefix match on error codes (e.g., "88a1"). device_id: Filter by device identifier. event_type: Filter by event type (command, response, error, etc.). top_k: Max results (default 20, max 50). Clamped to range [1, 50].

Returns: Formatted results with metadata and relevance scores.

Inputs: query, top_k, project, severity, +5 more. Returns: result

search_specsA

Search only specification and documentation files (filters out source code). Use when you need "what does the spec say about X" without code noise.

Args: query: Natural language search query project: Optional project name filter top_k: Number of results (default: 5, max: 20)

Returns: Relevant documentation chunks (specs, requirements, design docs, PDFs)

Inputs: query (required), top_k, project. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

Deploy Server

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4/5.0

Scored 2026-08-04 15:45 across 20 tools

Disambiguation4/5

Most tools have clearly distinct purposes: search_code vs search_specs vs search_logs are well-separated, and management tools (add_project, add_file, add_folder, add_pattern) have distinct scope. The only potential overlap is search_docs vs search_specs, where search_docs searches general documentation while search_specs filters to specifications/design docs, which could cause occasional misselection.

Naming Consistency5/5

All tools follow a consistent snake_case verb_noun pattern: search_*, add_*, remove_*, list_*, get_*, find_*, index_*, clear_*, compare_*. The verbs are descriptive and predictable, making it easy for an agent to guess tool names for actions.

Tool Count4/5

At 20 tools, the server is slightly heavy but each tool covers a distinct operation within the RAG domain: project management, file indexing, semantic search, code navigation, and log search. While more than the typical 15-tool threshold, the count is appropriate for the server's broad scope and each tool has a clear role.

Completeness4/5

The tool surface covers the full lifecycle: add/remove projects and files, search across docs/code/logs, retrieve document chunks, and list metadata. The only minor gap is that after clear_project_index there's no explicit reindex-all command, requiring users to re-add patterns or files manually, but this is a workaround rather than a dead end.

Maintenance

ActivityMaintained

ResponsivenessNo recent community issues

Resources

Related MCP Connectors

SynaplanA
io.github.metadist
Self-hosted AI workspace: semantic RAG document search, long-term memory, and AI chat via MCP.
RAG Systems Knowledge & Memory AI & Machine Learning
101921
https://web.synaplan.com/mcp
greadA
io.github.NitroRCr
An MCP server that gives your AI access to the source code and docs of all public github repos
Version Control Code Analysis Documentation Access
5365
https://api.gread.dev/mcp
ChromaDB Remote MCP Server
io.github.meloncafe
Remote ChromaDB vector database MCP server with streamable HTTP transport
Vector Databases RAG Systems Databases
212
https://meloncafe.github.io/chromadb-remote-mcp/mcp
mcpA
ai.llmse
Public MCP server for the LLM Search Engine
Search RAG Systems Developer Tools
102
https://llmse.ai/mcp

Related MCP Servers

mcp-context
RAG Systems Search
ericlimabr
F
license
Not graded
quality
D
maintenance
Local MCP server that provides semantic search (RAG) over code repositories, enabling AI clients like Claude and Gemini to access project context without manual re-upload.
Updated 6 months ago (2026-05-09 19:18 UTC)
-
ragi
RAG Systems Search
SusuTawar
A
license
A
quality
D
maintenance
Local-first RAG indexing and semantic search MCP server. Enables document retrieval and context-aware queries using local embedding models.
Updated 3 months ago (2026-08-13 16:07 UTC)
3
8 npm
MIT
ragtag-mcp
RAG Systems Vector Databases
weston-barger
A
license
Not graded
quality
D
maintenance
A local RAG MCP server that enables AI tools like Claude to search indexed codebases and documentation using vector search with Ollama models.
Updated a year ago (2025-09-15 04:00 UTC)
Apache 2.0
corpus-rag
RAG Systems Knowledge & Memory
monahand1023
A
license
Not graded
quality
B
maintenance
MCP server for local RAG over personal notes, PDFs, and documents, enabling plain-English querying and hybrid search with multi-hop context expansion.
Updated 19 days ago (2026-10-27 12:21 UTC)
MIT

pow-mcp-rag-new

Overview

License

Quick Start

PyPI (recommended for new users)

Local PyPI + uvx (recommended for development/publishing workflows)

Docker (alternative)

pip install (hosted index, once available)

Documentation

What gets indexed

MCP Tools

Log Indexing

Need Help?

Available Tools

Tool Schema Changelog

TDQS

Maintenance

Resources

Related MCP Connectors

Related MCP Servers

mcp-context

ragi

ragtag-mcp

corpus-rag