For coding agents

Deterministic tools, not vibes.

Gittensory ships an MCP server that exposes structured, schema-validated tools for plan, preflight, branch analysis, and packet prep. Use it as the base layer for Codex, Claude Desktop, Cursor, or your own agent runtime.

MCP client setup API reference

stdio + remote

Run the MCP as a local stdio process, or connect to the remote MCP at the Worker.

gittensory-mcp --stdio

npx -y @jsonbored/gittensory-mcp@latest --stdio

https://gittensory-api.aethereal.dev/mcp

http

Structured tool schemas

Every tool input/output is typed and validated. Agents get deterministic, machine-readable results.

· plan_next_work
· preflight_branch
· analyze_branch
· explain_blockers
· prepare_pr_packet

Init helpers

Print ready-to-paste config snippets for Codex, Claude Desktop, and Cursor.

gittensory-mcp doctor
gittensory-mcp status
gittensory-mcp init-client --print codex
gittensory-mcp init-client --print claude
gittensory-mcp init-client --print cursor

Workflow

The agent loop, made boring on purpose.

Predictable inputs, predictable outputs. The agent makes decisions; Gittensory makes those decisions explainable.

step 1

plan

Decision pack → repo targets + ranked next actions.

step 2

preflight

Metadata-only branch analysis with scenarios.

step 3

explain

Concrete, human-readable blockers and cleanup steps.

step 4

packet

Public-safe PR description ready to commit/push.

Agents never need a GitHub PAT or repo source. Inputs are metadata; outputs are structured. The MCP enforces both.

How it fits your agent

Drop-in tools, predictable shapes.

Three places Gittensory makes your agent's contribution loop boring on purpose.

Step 1 · Boot

Print client config

`init-client` writes the MCP config for Codex, Claude Desktop, or Cursor. No copy-paste required.

Step 2 · Plan

Deterministic ranked work

`plan_next_work` returns repo targets and ranked actions over the deterministic decision pack.

Step 3 · Ship

Packet, not prose

`prepare_pr_packet` returns a structured packet with labels and linked-issue lines ready to commit.

Wire it into your client

Set up MCP in your editor, then explore each tool in the live playground.

Client setup Open playground