Claude Code, Cursor, Codex: brilliant, and new to your company every single morning. They re-read your repo to rediscover what your team already knows. That is the context tax. Heidi ends it.
# Without Heidi
Agent reads 40 files to guess why the retry cap exists.
~180k tokens. The answer still is a guess.
# With Heidi
memory_search("retry cap decision")
→ "Cap fixed at 3 after the Meridian incident, 2025-11-02.
Contractual for two enterprise customers. Source: incident
review, legal thread #482."
~600 tokens. The answer has receipts.Code says what. It never says why. The why lives in meetings, incident reviews, Slack threads, and the heads of people who might leave. Heidi’s Cortex holds exactly that, with sources. Your agent asks one question and gets the institutional answer instead of an educated guess built from 40 files.
And it goes both ways: what your agents learn and decide can be written back as facts, so the next agent, and the next human, starts smarter.
# 1. Add Heidi as an MCP server (once per machine)
claude mcp add --transport http heidi https://mcp.falkster.ai/mcp \
--header "Authorization: Bearer $HEIDI_API_TOKEN"
# 2. That's it. Ask away.
claude "Why is the payment service split into two deployables?"# CLAUDE.md — add this to your repo
Before exploring the codebase for architectural or historical
questions, query the "heidi" MCP server first:
- memory_search: decisions, incidents, ownership, customer promises
- ask_heidi: anything that needs reasoning over company context
Heidi's answers carry sources. Prefer one Heidi call over
re-reading directories of code for "why" questions.Drop that into your repo’s CLAUDE.md and every session starts by asking the brain first. Your whole team gets it on the next pull.
Real prompts from real sessions, all answered from memory in one call: “Why did we pick speed over flexibility in the rendering engine?” “Which customers are contractually affected if I change the retry cap?” “Who owns the ingestion pipeline and what broke last time someone touched it?” “What did we promise Acme in the security review?”
Every answer comes with the source: the meeting, the thread, the document. Your agent can cite it in the PR description. Your reviewer can check it.
One token, one command, every Claude Code session in your company starts with institutional knowledge.
Get your API key