flikt-mcp

MCP server that connects Claude (Claude Code, Claude Desktop, or any MCP client) to your Flikt.AI account: browse plan-review results, dig into coordination conflicts, ask questions, export RFI packages, and — if you allow it — kick off reviews of uploaded plan sets.

Setup

1. Mint a token in the Flikt portal: Settings → API access → Create token. Default tokens are read-only. Check "Allow this token to start reviews" (and set a monthly page limit) only if you want Claude to be able to start reviews against your subscription's page allowance — a token can never charge your card; runs that aren't fully covered by your plan are refused.

2. Copy the token at creation — it is shown exactly once.

3. Add the server (Claude Code):

   claude mcp add flikt \
     --env FLIKT_API_TOKEN=flk_your_token_here \
     -- python -m flikt_mcp
   

   or in any MCP client config:

   {
     "mcpServers": {
       "flikt": {
         "command": "python",
         "args": ["-m", "flikt_mcp"],
         "env": { "FLIKT_API_TOKEN": "flk_your_token_here" }
       }
     }
   }
   

   Install first with pip install . (from this directory), or run via uvx --from <path-to-flikt-mcp> flikt-mcp.

FLIKT_API_BASE overrides the API host (default https://api.flikt.ai).

Tools

Spend safety

run_review rides your subscription's page allowance and nothing else. The Flikt API refuses (HTTP 402) any run that would require a payment — overage, per-project pricing, or spec add-ons — and tells you to finish in the portal instead. Per-token monthly page limits (set at mint time) bound what a single token can consume; revoking a token in Settings cuts access instantly.

Agentic review skill

skills/review-my-plan-set/SKILL.md packages the full loop — check the plan set, confirm the spend with you, start the review, poll, then deliver a top-risks summary and the RFI package. Point your Claude skills directory at it or copy it into .claude/skills/.

Development

pip install -e ".[dev]"
pytest