rag-tool-agent
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
RAG Tool Agent — MCP Server
Exposes the two tools behind the
rag-tool-agent-demo
project (retrieval-grounded Q&A over FordA dataset notes, and a safe
arithmetic calculator) as a Model Context
Protocol server, so any MCP-compatible
client (Claude Desktop, an MCP-aware agent harness, or a custom host) can
call them directly instead of only through that project's own CLI or through
the Flask HTTP wrapper in rag-tool-api.
What it does
The original agent does its own routing internally: a hand-rolled if/else decides "calculator" vs. "retrieval" vs. "direct answer" for each incoming question. MCP inverts that: the client (the LLM/agent host) does the routing, by reading each tool's name, description, and JSON-schema parameters and deciding which one to call. So this server exposes the two underlying capabilities as standalone, independently-callable MCP tools and lets any MCP client route to them itself. Tools become host-agnostic instead of hardwired into one agent's dispatch logic.
search_notes(query: str) -> str— retrieval-style lookup over a small inlined notes corpus about the FordA dataset, returned with the same[Grounded in N retrieved chunk(s) from ...]provenance suffix the original agent uses, so a client can tell a grounded answer from an ungrounded one. It uses a small keyword-overlap ranker rather than the original project's FAISS/embeddings index, since this project focuses on the MCP exposure layer.calculate(expression: str) -> str— arithmetic tool, restricted to+ - * / ()and numeric literals via anast-based safe evaluator (not a bareeval()on arbitrary input). It rejects both non-arithmetic input and injection attempts like__import__('os').system(...).Built with the official
mcpPython SDK (FastMCP), the same SDK Anthropic publishes for building MCP servers.
Related MCP server: rag-mcp
Running it
pip install -r requirements.txt
python server.pyThe server runs over stdio, the standard local transport MCP clients like Claude Desktop use to launch and talk to a server as a subprocess. To point Claude Desktop at it, add to its MCP server config:
{
"mcpServers": {
"rag-tool-agent": {
"command": "python",
"args": ["/absolute/path/to/server.py"]
}
}
}Tests
pytest tests/ -v11 tests, all passing. They call the tool functions directly (no MCP transport needed for unit-level coverage of the tool logic): grounded retrieval answers, the grounding-count contract, arithmetic correctness (including a division example matching the original agent's own documented example), rejection of non-arithmetic and injection input, division-by-zero handling, and tool self-registration with descriptions.
Results
Beyond the unit tests, I verified the server end-to-end with the real mcp
client SDK (ClientSession + stdio_client): spawning server.py as an
actual subprocess, completing the MCP initialize handshake, calling
list_tools(), and calling both tools over the real protocol:
TOOLS: ['search_notes', 'calculate']
calculate -> 1320 / (3601 + 1320) = 0.2682381629750051
search_notes -> The FordA dataset is a univariate time-series
classification dataset... [Grounded in 3 retrieved chunk(s) from
forda_dataset_notes.md]
calculate(injection) -> Error: could not evaluate "__import__('os')" ...This confirms the server handles the handshake, tool discovery, and tool invocation over real MCP.
Related projects
rag-tool-agent-demo— the original CLI agent: LangChain, FAISS, Ollama, LCEL retrieval chain, calculator tool, LLM-driven routing.rag-tool-api— that agent wrapped as a Flask HTTP API, containerized (Docker, multi-stage build, non-root user, health check), verified with real HTTP requests against a running container.This project — the same underlying capabilities (retrieval, calculator) exposed over MCP instead of HTTP, so they're callable by any MCP host rather than only by a REST client.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Salesforce-grounded retrieval, diagnoses, and a vetted-Force marketplace for MCP clients.
Free OpenAI-compatible inference with signed provenance receipts and 3 focused MCP tools.
Find, vet, and run MCP tools through a secure audited gateway with prompt-injection risk scoring
Agentic search over your Dewey document collections from any MCP-compatible client.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables document search, read-only SQL querying, arithmetic calculation, and corpus introspection through MCP tools, allowing Claude to dynamically discover and chain these tools together to answer multi-step questions.-
- FlicenseNot gradedqualityDmaintenanceEnables querying local documents via retrieval-augmented generation, along with current UTC datetime and safe arithmetic evaluation through MCP tools over HTTP.-
- AlicenseAqualityCmaintenanceProvides a document store, web search, and safe calculator as MCP tools, enabling both offline and LLM-driven clients to discover and call them over the Model Context Protocol.6MIT
- FlicenseNot gradedqualityCmaintenanceEnables an LLM to perform safe arithmetic calculations and web searches through a standardized MCP interface, returning results to generate informed responses.-