Misikiri/Levelupp/MCP

Your domain's real rules, one tool call away.

Levelupp MCP is the delivery component of the Levelupp platform — a Model Context Protocol server that injects your own expert-validated knowledge into every relevant prompt. Connect once, then forget it's there.

30-second setup · Streamable HTTP · API key or OAuth 2.0 + PKCE
claude-code — connected to levelupp
> write the logic that decides when an action can be auto-approved
⏺ levelupp · enrich_prompt
---LEVELUPP EXPERT CORE---
[rule_017] Auto-approval needs two independent checks
above threshold — one high-confidence signal alone
is not sufficient.
[edge_case_006] Records changed in the last 24h →
route to manual review.
---END LEVELUPP EXPERT CORE---
objects_applied: 2  ·  confidence: high
Quick start

Connect once. Forget it's there.

Levelupp speaks standard Model Context Protocol over streamable HTTP. Add it to your client, and your AI calls it automatically whenever a prompt touches your domain — no workflow change.

~/.claude.json · mcpServers
{
  "mcpServers": {
    "levelupp": {
      "url": "https://levelupp.misikiri.com/mcp",
      "headers": { "X-API-Key": "lup_your_key_here" }
    }
  }
}

That's the whole integration. Your API key identifies your workspace and you as a builder — issued by your workspace admin.

Claude Desktop · Settings → Connectors → Add custom connector
# Add the Levelupp connector URL — sign-in happens in the browser
https://levelupp.misikiri.com/mcp

# Levelupp supports OAuth 2.0 + PKCE — no key to paste,
# log in with your builder account and you're connected.
Any MCP-compatible client · streamable HTTP
# Endpoint (streamable HTTP transport)
https://levelupp.misikiri.com/mcp

# Auth — either header-key or OAuth 2.0 + PKCE
X-API-Key: lup_{workspace}_{builder}_{secret}

The enrichment contract

A contract, not a vibe

Enriched prompts carry domain context in an exact, machine-parseable delimiter block. If nothing matches, your prompt passes through untouched with objects_applied: [] and low confidence — Levelupp never pads, never invents.

  • Deterministic output schema: enriched_prompt, objects_applied, confidence
  • Only certified knowledge can ever appear in the block — see how certification works
  • Typical enrichment adds ~100ms — keyword-matched, not a model round-trip
enrich_prompt → response
{
  "enriched_prompt": "<your prompt>

---LEVELUPP EXPERT CORE---
[certified domain context]
---END LEVELUPP EXPERT CORE---",
  "objects_applied": ["rule_017", "edge_case_006"],
  "confidence": "high"
}
Tool reference

Everything your agent can call.

Nine MCP tools across enrichment, expert sessions, and Validated AIDLC.

Core enrich_prompt

The flagship. Matches a prompt against your certified knowledge and returns it enriched with the Expert Core block, applied object IDs, and a confidence rating.

Core consult

Ask the Expert Core a direct domain question — including stage-aware Validated AIDLC guidance — without wrapping it in a build prompt.

Core get_workspace_knowledge

Browse the certified knowledge available to your workspace: shared baseline plus your private overrides and additions.

Sessions create_session

Spin up a knowledge-capture session against a Business Objective straight from your coding agent.

Sessions get_session_context

Pull the accumulated context of an existing session into your current conversation.

Sessions save_session_context

Persist what your agent learned back to the session — context survives across conversations and machines.

Validated AIDLC aidlc_status

Where is this project in the lifecycle? Registration state, stage, riders, and what's next.

Validated AIDLC get_aidlc_artifacts

Fetch the versioned governance artifacts written into your repo — the CODEOWNERS gate and CI eval gate.

Validated AIDLC get_bo_summary

Summarise a Business Objective: participants, sessions, and the knowledge produced against it.

FAQ

Developer questions, answered.

What exactly is an MCP server?

The Model Context Protocol is an open standard (originated by Anthropic, now Linux Foundation infrastructure) that lets AI assistants call external tools and data sources. Levelupp implements it over streamable HTTP, so clients like Claude Code and Claude Desktop can pull certified domain knowledge into a conversation automatically — no plugins, no copy-paste.

Does this slow my coding agent down?

No. Enrichment is keyword-matched against a pre-merged in-memory index — typically ~100ms, not an LLM round-trip. If nothing matches, your prompt passes through unchanged.

What do I need to get started?

A builder key (issued by your workspace admin) or a builder login for the OAuth flow, plus the three-line MCP config above. Setup is around 30 seconds.

Which clients are supported?

Anything that speaks MCP over streamable HTTP: Claude Code, Claude Desktop (as a custom connector with OAuth), and any other MCP-compatible client using the X-API-Key header.

When does my AI actually call Levelupp?

The tool descriptions instruct the model to call Levelupp before responding to prompts that touch your domain — the areas your certified knowledge covers. Outside those areas, your prompts are untouched.

Where does the knowledge come from?

From you. Levelupp reads what your tools already reach and adds its own research to work out what needs validating; your domain experts certify it, and only certified objects are ever served. The full governance story — certification, workspace isolation, Validated AIDLC, provenance — is on the Levelupp product page.

Get a builder key and connect in seconds.

Levelupp MCP comes with a Levelupp workspace. Request access for your team, or read about the platform behind the server.

Copied to clipboard