Skip to main content
Glama
Liminate Authoring MCP

Liminate Authoring MCP

Liminate Authoring MCP

Help your AI agent author the Agreements that govern it.

Part of the Liminate family — trust infrastructure for autonomous AI, built on a 61-word prose-as-syntax language.

liminate-mcp is a standalone MCP server that lets AI agents validate, explain, draft, and test Liminate Agreements — and translate compliance documents into Agreement drafts via the Liminate platform. It never installs an Agreement. Installation onto an enforcement surface (like ~/.seshat/agreement.limn) is always a human or Seshat-governed action, never this server's.

Built by Liminate

Repo

What it does

seshat-app

Local agent harness. Enforces Agreements deny-by-default.

liminate

The language and interpreter. 61 words, deterministic execution, domain packs.

liminate-invariant

Semantic verification harness. Deterministic claim-verification correction loop.

← this repo

liminate-mcp

Authoring MCP server. Helps an agent write the Agreements that govern it.

→ liminate.dev

Related MCP server: doc-version-mcp

Tools

Five tools run entirely on your machine, against the local liminate interpreter — no network calls, no account required:

Tool

Cost

What it does

validate_agreement

free

Parse and semantically check an Agreement. Errors vs. unbound evidence references, annotated separately.

explain_clause

free

Plain-English explanation per clause: verb, condition, rationale, temporal window, exception.

draft_agreement

free

Assemble an Agreement from structured rules, or return a vendored template. Always validates its own output.

test_agreement

free

Run an Agreement against sample evidence; returns a decision (blocked / incomplete / needs_review / passed).

list_templates

free

The five vendored Agreement templates: EDGAR revenue check, quarterly earnings brief, publication review, student assignment review, vendor invoice review.

One tool calls the platform:

Tool

Cost

What it does

translate

paid

Converts a compliance document (text/markdown/url) into an Agreement draft via the liminate.dev Translate API. Requires LIMINATE_API_KEY (liminate.dev/keys); bills against that account. Always returns a fidelity manifest — review it before installing the draft anywhere.

Install

Requires Python 3.11+.

pip install liminate-mcp

Or from source:

git clone https://github.com/rmichaelthomas/liminate-mcp.git
cd liminate-mcp
pip install -e ".[dev]"
pytest

MCP client config

{
  "mcpServers": {
    "liminate-authoring": {
      "command": "liminate-mcp"
    }
  }
}

Set LIMINATE_API_KEY in your environment (or the MCP client's env block) to use translate. The five local tools work with no key and no network access.

License

Apache 2.0. See LICENSE.


The interpreter is the trust root. Every draft is validated before it's handed back — never installed, never assumed.

Available Tools

6 tools
draft_agreementA

Assemble a Liminate Agreement from structured rules (verb/condition/ reason/unless/starting/until), or return a vendored template by template_id. Always validates the drafted source against the interpreter before returning — never claims a draft is valid without confirming it.

Inputs: about, rules, template_id.

explain_clauseB

Explain each clause of a Liminate Agreement in plain English: verb, condition, rationale, temporal window (active/expired/future/unbounded), and any unless-exception.

Inputs: source (required).

list_templatesA

List the four vendored Liminate Agreement templates (id, name, purpose, agreement source, and sample evidence).

Inputs: no named parameters.

test_agreementB

Run a Liminate Agreement against sample evidence and return the decision: blocked, incomplete (missing evidence), needs_review, or passed.

Inputs: source (required), evidence (required).

translateA

PAID — convert a compliance document (source_kind: text/markdown/url) into a Liminate Agreement draft via the liminate.dev Translate API. Requires LIMINATE_API_KEY (get one at liminate.dev/keys); bills against that account. Always present the returned fidelity manifest to the user and warn them to review the draft before installing it — it controls what agents can do on their machine.

Inputs: source_kind (required), url, text.

validate_agreementA

Parse and semantically check a Liminate Agreement. Returns parse/semantic errors and a per-rule deontic analysis (permit/forbid/require/define), annotating unbound evidence references separately from genuine errors.

Inputs: source (required).

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool has a distinct purpose: drafting, explaining, listing templates, testing, translating, and validating. There is no overlap or ambiguity in their functions.

    Naming Consistency4/5

    Most tools follow a verb_noun pattern (e.g., draft_agreement, explain_clause), but 'translate' is a single verb without a noun, deviating slightly from the convention.

    Tool Count5/5

    With 6 tools covering creation, validation, testing, explanation, template listing, and document translation, the count is well-scoped for an authoring server.

    Completeness3/5

    The set covers core authoring operations but lacks tools for retrieving or managing existing agreements (e.g., list_agreements, get_agreement), leaving a notable gap in the lifecycle.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      Contract and wills legal tools for AI agents: statute-cited requirements per contract type and US state, non-compete enforceability for all 50 states + DC, intake schemas, async multi-agent contract drafting with free previews, and contract risk analysis.
      8
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables local-first document versioning with content-addressable storage, word-level AST diffs, and AI-compliance auditing for LLM agents.
      6
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents and builders to run local, audit-traceable governance assessments, including an A³ scorecard, ISO/IEC 42001 and NIST evidence templates, compliance checklists, passport lookup, and sensitive-text screening.
      MIT