Tetrees MCP

Tetrees Agent EX

Public integration kit for using, growing, buying, and publishing Tetrees AI Packs through the hosted Tetrees Agent MCP and API.

[Tetrees Agent selecting useful intelligence from a holographic shelf]

Explore Tetrees AI Packs  ·  Connect through MCP  ·  Publish an AI Pack

This repository is the public @tetrees/mcp product (ai.tetrees/mcp): the stdio MCP server, clients, configuration examples, and safe extension samples. It does not contain Tetrees Agent internals, private Pack intelligence, Agent AVCP scoring, service prompts, storage, moderation, or exchange infrastructure.

What you can build

Tetrees-hosted work needs no local GPU. Model-funded runs are quoted in Tetrees Points before execution; an owned Pack can also use a supported provider key supplied to the local MCP process.

Quick start: hosted MCP

Requirements: Node.js 20+ and a revocable API token from Tetrees Account.

Add this to your MCP client's configuration:

{  "mcpServers": {    "tetrees-ai": {      "command": "npx",      "args": ["-y", "https://ex.tetrees.ai/pkg/tetrees-mcp.tgz?v=2.2.1"],      "env": {        "TETREES_API_URL": "https://ex.tetrees.ai/api",        "TETREES_TOKEN": "<revocable-account-token>"      }    }  }}

Restart the MCP client, then ask it to:

Search Tetrees AI Packs for repository issue triage. Show the public Agent AVCPreport and runtime profile for the best match, then quote an 8k/2k Point-fundedrun. Do not run it until I approve the quoted maximum.

Do not put provider keys in a Pack, prompt, source file, or shared MCP config. Optional BYOK keys belong only in the local MCP process environment. See examples/mcp/mcp-config.jsonc.

Inspect the MCP contract

npm installexport TETREES_TOKEN="txk_..."npm run mcp:inspect

The inspector launches the official hosted MCP package, lists its public tools and resources, checks the expected buyer/builder/seller surface, and exits without spending Points or changing account state.

The v=2.2.1 suffix is intentional: it prevents npx from reusing an older MCP tarball cached under the same URL. Change it only when Tetrees publishes a newer MCP contract. The npm identity for the same package is @tetrees/[email protected].

Public API CLI

Copy .env.example values into your shell; this project deliberately does not auto-load secret files.

export TETREES_TOKEN="txk_..."
# Read-only callsnpm run tetrees -- statusnpm run tetrees -- catalog "research"npm run tetrees -- modelsnpm run tetrees -- skillsnpm run tetrees -- ownednpm run tetrees -- runtime <pack-id>npm run tetrees -- report <pack-id>
# Quote first. This never starts a model request.npm run tetrees -- quote <pack-id> <model-id>
# A Point-spending run needs the exact confirmation flag.npm run tetrees -- run <pack-id> <model-id> "Your task" --confirm-spend

The CLI redacts secret-shaped response fields and reports Tetrees error codes, HTTP status, and Retry-After. It does not retry a mutation automatically.

Buyer flow

1. Search for a Pack.
2. Inspect its public Agent AVCP report and runtime profile.
3. Quote the exact model, token ceiling, hosted skills, and growth checkpoint.
4. Run a Point-funded stateless preview. Unused reserved Points return.
5. Acquire a free Pack or buy a paid Pack on Tetrees EX.
6. Ownership unlocks download, BYOK, saved growth, and client extensions.

Run the documented read-only example:

node examples/buyer/discover-and-quote.mjs "incident response"

Builder flow: add capabilities without giving them to Tetrees

A published Pack may declare client extensions that passed Agent AVCP. The actual binding and credentials remain on the user's machine.

Supported binding patterns:

• https_api: fixed HTTPS base URL, declared methods, environment-backed headers.
• mcp: a nested MCP server started without a shell and mapped to declared tool names.
• local_skill: a no-shell child process receiving one JSON request and returning one JSON result.

Start with:

• examples/extensions/client-extensions.example.json
• examples/extensions/local-skill/issue-summary.mjs
• docs/BUILDER_GUIDE.md

Tetrees hosting never inherits these local permissions. A write method needs the exact per-call phrase APPROVE extensionId.methodId.

Seller flow

The supported publisher path is intentionally contract-driven:

# Read-only readiness/report inspectionnode examples/seller/check-readiness.mjs <pack-id>
# Upload one higher immutable version. Requires explicit confirmation.node examples/seller/upload-pack.mjs <pack-id> 1.1.0 ./release.taip --confirm-upload

Then use the hosted MCP tools or Tetrees API reference to submit, quote Agent AVCP, run it after confirming the exact Point amount, inspect the private report, and publish only the passing digest.

See docs/SELLER_GUIDE.md and examples/seller/pack-listing.example.json.

Stable public references

• MCP guide
• API documentation
• OpenAPI 3.1
• Postman collection
• AI Pack catalogue

Security and product boundary

This repository does not contain:

• Tetrees Agent implementation, private Pack contents, memory representation, or growth logic;
• Agent AVCP scoring rules, internal evaluation fixtures, service prompts, or moderation logic;
• model-provider credentials, pricing margins, signing keys, storage details, databases, admin APIs, deployment code, or production configuration.

The examples operate only through the documented public, authenticated contract. Review SECURITY.md and docs/PUBLIC_BOUNDARY.md before contributing.

Verify this repository

npm testnpm run check:public-boundarynpm pack --dry-run

License

The client examples and documentation in this repository are MIT licensed. Tetrees, Tetrees Agent, Tetrees AI Pack, TAIP, and Agent AVCP names and service marks remain the property of Tetrees EX. The license does not grant access to, or rights in, the hosted service or private Pack intelligence.