DIT
Do It Together

MCP Server

Evidence-backed playbooks for AI agents.

Connect your AI agent to a catalog of 0 canonical paths synthesized from 0 real engineering arcs. Every playbook is grounded in actual successful state transitions.

Browse catalog
Canonical Paths
0

Synthesized playbooks

Clusters
0

Problem domains

Source Arcs
0

Real state transitions

MCP Console
Connected
$ npx @doittogether/mcp-server
[info] DIT MCP Server running on stdio
→ search_paths { query: "build errors" }
✓ Found 8 paths in web-app-debugging cluster
→ get_path { id: "resolve-build-time-syntax-and-reference-errors" }
✓ Loaded canonical path (5 steps, 3 success criteria)
▌
Paths
0
Clusters
0
Arcs
0

Knowledge Pipeline

Session Logs
Real engineering work
Arcs
State transitions
Clusters
Grouped by intent
Canonical Paths
Golden playbooks
harvest_sessions extracts arcs
search_paths / get_path query catalog
Performance BenchmarksComing Soon

We're measuring MCP-enabled vs baseline agent performance across SWE-bench and other engineering benchmarks. Results will be published here once validated.

—
Pass rate delta
—
Time saved
—
Retries avoided

Quick Start

Install from npm and connect your MCP-compatible client. No cloning required.

npx @doittogether/mcp-server

Requires Node.js 20+. Package: @doittogether/mcp-server

Why Evidence-Backed?

Grounded in reality
Every canonical path derives from real engineering sessions, not speculation.
Success signals included
Each arc captures explicit evidence that the approach worked.
Continuously growing
Contribute session logs to expand the knowledge base.

Tool Surface

Catalog

0 canonical paths across 0 engineering domains.

  • • search_paths
  • • get_path

Harvest

Extract arcs from session logs to grow the knowledge base.

  • • harvest_sessions

Capture

Rich system probes for grounded agent reasoning.

  • • capture_state_full

How It Works

Step 1
Connect MCP

Attach the DIT server to your AI agent (Claude, Cursor, etc.).

Step 2
Search catalog

Find canonical paths for your problem domain.

Step 3
Get context

Capture system state for grounded reasoning.

Step 4
Contribute

Your session logs can become new arcs.

Example Agent Prompt

Search the catalog and retrieve a canonical path for your problem.

"Use MCP server 'dit'.
1) search_paths { query: 'debugging build errors' }
2) get_path { id: 'resolve-build-time-syntax-and-reference-errors' }
3) Follow the canonical path steps to solve the problem."

Contributing Arcs

  • •Session logs from your AI agent contain valuable state transitions.
  • •Use harvest_sessions to extract arcs from logs.
  • •Arcs get clustered and synthesized into new canonical paths.
Every contribution helps grow the knowledge base for the community.

Data & Privacy

Local-first
  • • MCP server runs entirely on your machine.
  • • Catalog data is bundled locally.
  • • No data sent externally by default.
Optional contribution
  • • Opt-in to share harvested arcs.
  • • Review before submission.
  • • Helps grow the canonical path catalog.

Troubleshooting

  • • Ensure Node.js 20+ is installed (node --version).
  • • Try running npx @doittogether/mcp-server directly to verify it starts.
  • • Verify your client supports MCP stdio transport.
  • • Restart Claude Desktop/Code after updating the config file.
Questions? Browse the catalog or check the API documentation.