RETRIEVAL AND GROUNDING / Vectara
Vectara MCP Server
Runs Vectara RAG queries and semantic search over named corpora, and calls Vectara's hallucination-correction and factual-consistency APIs, from an MCP client.
“Vectara-MCP provides any agentic application with access to fast, reliable RAG with reduced hallucination, powered by Vectara's Trusted RAG platform, through the MCP protocol.”
01 / THE REASONING
Why this made the selection.
- Published in Vectara's own repository under Apache-2.0; the README at this commit lists six tools, including correct_hallucinations and eval_factual_consistency, which check generated text against supplied source documents.
- The README states that HTTP transport is the default with bearer-token authentication on and a 100 requests/minute default rate limit, and that stdio must be enabled explicitly with --stdio.
02 / THE REVIEW RECORD
What we actually inspected.
Source review has boundaries.
A clear record is more useful than a “safe” badge.
Material inspected
- README.md at 2b5f109e1fca388dea0931f32749e575a7cc2c1d
- LICENSE at the same commit
- pyproject.toml at the same commit
Our findings
- python -m vectara_mcp starts HTTP on 127.0.0.1:8000 with authentication enabled; --no-auth disables it and the README marks that development only.
- VECTARA_ALLOWED_ORIGINS configures CORS and VECTARA_AUTHORIZED_TOKENS adds accepted bearer tokens for the HTTP transport.
- pyproject.toml names vectara-mcp 0.2.1, Python >=3.11; it has no license field, and the LICENSE file is Apache-2.0.
Not established by this review
- pip install
- Any Vectara query
- The hallucination-correction and factual-consistency tools
The review applies to the material and revision named here. A newer upstream release can change its behavior.
03 / PUT IT TO WORK
Connect a server deliberately.
DOCUMENTED COMMAND
pip install vectara-mcp Copying does not execute this command. It may retrieve a newer version than the reviewed source.
- Install the package with the README command above. It was not run.
- For Claude Desktop, add the README's stdio block with VECTARA_API_KEY in env. For HTTP, keep authentication on and set VECTARA_ALLOWED_ORIGINS. Do not commit the key.
Before you start
- Python 3.11 or later
- A Vectara API key and corpus keys
Documented configuration example
Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.
{
"mcpServers": {
"Vectara": {
"command": "python",
"args": [
"-m",
"vectara_mcp",
"--stdio"
],
"env": {
"VECTARA_API_KEY": "your-api-key"
}
}
}
} 04 / FOLLOW THE EVIDENCE
The source trail.
Our notes are separate from the original resource.
Check upstream before adopting a new version.
- Vectara MCP README ↗Checked
https://github.com/vectara/vectara-mcp/blob/2b5f109e1fca388dea0931f32749e575a7cc2c1d/README.md
Supports: Tools, Transport defaults and authentication, Claude Desktop configuration, Default generation preset
- Vectara MCP LICENSE ↗Checked
https://github.com/vectara/vectara-mcp/blob/2b5f109e1fca388dea0931f32749e575a7cc2c1d/LICENSE
Supports: Apache License 2.0
- Vectara MCP pyproject.toml ↗Checked
https://github.com/vectara/vectara-mcp/blob/2b5f109e1fca388dea0931f32749e575a7cc2c1d/pyproject.toml
Supports: Package vectara-mcp 0.2.1, Python >=3.11