Skip to main content
Glama

ai-company-mcp

A thin, protocol-first governance layer for AI virtual companies, exposed as an MCP server.

CI

Status: SDD complete — Converged (round 2, 2026-11-17): all 12 AT-* tests green, first CI run 7/7 jobs green across the 6-cell matrix, community checklist complete. Run npm install, npm test; see the convergence report. Read the spec before proposing changes. 简体中文说明

What this is

An MCP server that turns a virtual company's governance — department routing, approval gates, audit records, and specification contracts — into standard MCP primitives (tools, resources, prompts). Any MCP-compatible host can use it. No agent runtime, no UI, no database, no LLM calls, no vendor lock-in.

The AI agent ecosystem has three layers. This project sits between them:

Layer

Concern

Examples

This project

Connectivity

How agents call tools

MCP, A2A

Speaks MCP 2026-07-28 (primary target) with negotiation for all published historical versions

Orchestration

How tasks and state flow

LangGraph, CrewAI, Agents SDK

Not an orchestrator

Organization

Who reports to whom, who approves

Paperclip, The Agency, trinity

A composable governance protocol layer

Complementary, not a replacement. Use The Agency for role personas, Paperclip for a full control plane, LangGraph for orchestration. Use this project when a host needs the governance semantics themselves, without adopting a platform.

Related MCP server: org-as-code MCP Server

Why another project

Competitor research (snapshot 2026-11-16, verified against the GitHub REST API) is in docs/research/competitive-analysis.md. The short version: the organization layer is either a heavy self-hosted app or a folder of Markdown personas. Neither ships a lightweight, testable, protocol-level governance contract. SDD + ATDD are first-class here: the spec is the contract, and company.spec.validate will make that contract callable.

Repository map

Path

Purpose

CONSTITUTION.md

Non-negotiable project principles (SDD Step 0)

docs/development-method.md

SDD phases, ATDD red-first loop, EARS/GWT grammar, traceability

docs/research/competitive-analysis.md

Competitor landscape and product decisions D1–D7

specs/001-core-mcp-server/spec.md

SDD Step 1 deliverable — feature specification

specs/001-core-mcp-server/plan.md

SDD Step 3 deliverable — implementation plan (architecture, decisions, risks)

specs/001-core-mcp-server/tasks.md

SDD Step 4 deliverable — task breakdown, each task naming its AT-*

specs/001-core-mcp-server/convergence-report.md

SDD Step 6 deliverable — convergence report (traceability, SC evidence, pending items)

scripts/check_repo.py

CI contract gate: community files, encoding, FR→AT traceability

CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md, SUPPORT.md

GitHub community health files

.github/ISSUE_TEMPLATE/, .github/PULL_REQUEST_TEMPLATE.md

Contribution templates

.github/CODEOWNERS

Code ownership (maintainer @JohnSmithfan)

src/, test/

v0.1 implementation and the executable AT-001..AT-012 suite (npm test)

scripts/bench-cold-start.mjs

SC-001 cold-start benchmark (npm run build && npm run bench)

scripts/e2e-stdio-client.mjs

End-to-end stdio validation over a real process boundary (npm run build && npm run e2e)

.vscode/mcp.json

One-click host registration for VS Code's MCP client (run npm run build first)

Run and validate (Node 20/22 LTS):

npm install && npm test        # 65 test cases covering AT-001..AT-012
npm run build && npm run bench # SC-001 cold-start benchmark (real subprocess, p95 < 200 ms)
npm run build && npm run e2e   # end-to-end stdio validation over a real process boundary

Development workflow

Spec-Kit-compatible SDD phases:

constitution -> specify -> clarify -> plan -> tasks -> implement -> converge
                                                                    ^ you are here

Every functional requirement FR-* maps to at least one acceptance scenario AT-*. Each AT-* must exist as a failing executable test before implementation. See docs/development-method.md.

Run the repository contract gate locally:

python -B scripts/check_repo.py --root .

No runtime, package manager, or credentials are needed for this step.

Before publishing to GitHub

The repository is published as JohnSmithfan/ai-company-mcp; the maintainer contact channel (GitHub private contact, @JohnSmithfan) and .github/CODEOWNERS were settled on 2026-11-16 (see the spec's Clarified Decisions). Checklist:

  • Fill the maintainer contact channel in CODE_OF_CONDUCT.md (2026-11-16).

  • Add .github/CODEOWNERS with the real maintainer handle (2026-11-16).

  • Enable GitHub Discussions (referenced by SUPPORT.md) and private vulnerability reporting (referenced by SECURITY.md) (2026-11-17).

  • Add CI/status badges after the first successful workflow run (2026-11-17).

GitHub community profile checklist requires: README.md, LICENSE, CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md, valid issue templates and a pull request template. SUPPORT.md and CHANGELOG.md are included as well. LICENSE cannot be inherited from a default .github repository — it lives here.

License

GNU General Public License v3.0 or later (SPDX: GPL-3.0-or-later). The verbatim FSF text of version 3 is in LICENSE. You may use and redistribute this project under GPLv3 or, at your option, any later version of the GNU GPL published by the Free Software Foundation.

GPLv3 is a strong copyleft license. In practical terms:

  • If you distribute a modified or derived version, the whole derived work must be licensed under GPLv3 and you must provide its complete corresponding source code.

  • Contributions are accepted under the same terms (inbound = outbound); there is no CLA or copyright assignment. See CONTRIBUTING.md.

  • A client that merely talks to this server over MCP is not automatically covered by this license. Whether a specific integration forms a single derived work is a legal question this project does not decide; seek your own advice before shipping a combined distribution.

The "How to Apply These Terms" boilerplate at the end of LICENSE is the FSF's generic template. This section states the project's actual grant: the or later option is granted here, not by editing LICENSE (the GPL forbids changing the license document). GitHub's community profile detects the GPLv3 family from the unmodified LICENSE; the SPDX expression above is authoritative for the version option.

Documents under docs/ and specs/ carry AI-generation disclosures and audit trace IDs; they require human review before being treated as decisions.

Tool Schema Changelog

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

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      Security-enforcing MCP proxy that sits between an AI agent and any number of downstream MCP servers, intercepting every tool call through a capability-token policy gateway that can allow, deny, or escalate to human approval before the call reaches any real tool. It also exposes built-in operator tools for approval workflows, audit trail queries, token management, voice/HUD output, and hierarchical
      21
      16
      Apache 2.0
    • A
      license
      B
      quality
      B
      maintenance
      A governance proxy for AI tools — every MCP/agent tool call is policy-gated, secret-redacted, and written to a hash-chained, offline-verifiable audit trail.
      13
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Exposes TELOS governance primitives—action scoring, receipt verification, Purpose Anchor inspection, audit-chain queries, and CCRS counterfactual replay—as MCP tools, resources, and prompts for any MCP-compatible client.
      5
      Apache 2.0