Skip to main content

Model Context Protocol server for Touchstone: hallucination detection for LLM outputs without calling another LLM. Wraps the calibrated Verifier and the raw measurement orchestrator from the clarethium-touchstone reference implementation as MCP tools.

Project description

touchstone-mcp

Model Context Protocol server for Touchstone: hallucination detection for LLM outputs without calling another LLM.

touchstone-mcp exposes the calibrated Verifier and the raw measurement orchestrator from the clarethium-touchstone reference implementation as MCP tools. Any MCP host (Claude Desktop, Claude Code, Cursor, custom) can attach it as a stdio MCP server.

Install

pip install touchstone-mcp

fastmcp and clarethium-touchstone install transitively. The touchstone-mcp console script is registered automatically and runs on the stdio transport by default.

Host config

Drop this into your MCP host config:

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

For Claude Code: claude mcp add touchstone touchstone-mcp.

Tools exposed

  • verify — calibrated (text, source) hallucination probability with scope classification, signal breakdown, and span-level localization.
  • measure — raw multi-layer Touchstone output (all eleven Section 5 measurement layers).
  • assess_derivation_regime — Layer 11 standalone regime classifier.
  • list_modes — enumerate the four Verifier modes with their required inputs.

See the host wiring guide for full tool reference, scope semantics, and threshold guidance.

Programmatic use

from touchstone_mcp import build_server

server = build_server()
server.run()                 # stdio transport

Version pairing

touchstone-mcp depends on clarethium-touchstone>=0.2.0. The MCP server is a thin wrapper around the library's public surface; measurement semantics and calibration coefficients are defined and tested in the library, not here.

License

Apache-2.0. See LICENSE.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

touchstone_mcp-0.1.0.tar.gz (12.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

touchstone_mcp-0.1.0-py3-none-any.whl (11.2 kB view details)

Uploaded Python 3

File details

Details for the file touchstone_mcp-0.1.0.tar.gz.

File metadata

  • Download URL: touchstone_mcp-0.1.0.tar.gz
  • Upload date:
  • Size: 12.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for touchstone_mcp-0.1.0.tar.gz
Algorithm Hash digest
SHA256 0b0769528e026fa7736139b5ed8c9700613e9be54f9680d5dee0014145d9cb4f
MD5 012af85f08d649ed4da76ac6260488cf
BLAKE2b-256 ee1980c9f23f4a8005191a758f30123fa5885eaa3b53aa0c48c3382f2b5ea739

See more details on using hashes here.

File details

Details for the file touchstone_mcp-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: touchstone_mcp-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 11.2 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for touchstone_mcp-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 8be5cda84b82fe320ba55bb0ccdd9663251b92c596d920834059d0c4348c6be3
MD5 e6eb61a5fb33d92229daed7e250d01c2
BLAKE2b-256 86e5a2b132306d11cc4ae813d16a6c39c21aa38689ea07775ae2eef3e248ac8e

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page