Entelligence

@Entelligence add the github.com/acme/api-gateway repository so we can query its codebase in future prompts

Repository added successfully. The api-gateway codebase is now indexed and available for queries. You can ask questions about its architecture, specific files, or implementation patterns. Initial indexing may take 2-3 minutes depending on repository size; subsequent queries will be faster as the system builds its understanding of the code structure.

This example shows the setup step required before querying code. The MCP indexes repositories on-demand, so there's a one-time delay per repo. No authentication is required, but you'll need valid repository URLs. Once added, the repository remains available for all team members in the Space.

@Entelligence where in our api-gateway repo do we handle rate limiting, and what's the current strategy?

Rate limiting is implemented in `src/middleware/rateLimit.ts` using a token bucket algorithm. The current strategy allows 100 requests per minute per API key, with burst capacity of 20 requests. Configuration is loaded from `config/limits.json` and can be overridden per endpoint using the `@RateLimit` decorator. The middleware integrates with Redis for distributed rate tracking across instances.

This demonstrates the MCP's ability to answer architectural questions by searching indexed code. Responses synthesize information across multiple files. Accuracy depends on how recently the repository was indexed—code changes after indexing won't appear until you re-add the repository. Best for understanding existing patterns rather than tracking live changes.

@Entelligence compare how authentication is handled in our api-gateway versus the legacy-auth-service repository, and flag any security inconsistencies

Both repositories use JWT tokens, but with notable differences: api-gateway validates tokens on every request using RS256 signatures and checks expiration + issuer claims. legacy-auth-service uses HS256 (symmetric keys) and only validates expiration. Security concern: legacy-auth-service stores the signing secret in a config file committed to the repository (line 47 of `auth/jwt.js`), while api-gateway correctly loads it from environment variables. Recommendation: migrate legacy-auth-service to RS256 and externalize secrets before deprecating.

This example shows cross-repository analysis, useful for migration planning or security audits. The MCP can compare patterns across codebases you've added. Response quality improves when you provide context about what you're looking for. Note that it analyzes static code—it won't catch runtime-only issues or test actual authentication flows.

When Entelligence works for ramping new engineers on unfamiliar repos

A 6-person engineering team brings on two junior devs who need to understand a 3-year-old Rails monolith with sparse documentation. Entelligence wins here because the repository agent answers context-specific questions without requiring the seniors to stop and explain every pattern. The new hires ask "why does this service call retry three times" and get answers grounded in the actual code, not generic Stack Overflow threads. The threshold: if your codebase changes daily and the agent's index can't keep up, answers lag behind reality and juniors learn outdated patterns. Best for stable codebases where the question volume justifies adding the repo once and querying it for weeks.

Entelligence speeds up code audits when evaluating a potential acquisition

A 12-person startup's CTO needs to assess an acquisition target's codebase in 72 hours before the term sheet expires. Entelligence handles this because the repository agent surfaces architectural decisions and technical debt patterns faster than manual grep-and-read. The CTO asks "how is authentication handled across services" and "where are database migrations managed" and gets synthesized answers instead of spending hours tracing imports. The trade-off: this only works if the target grants repo access upfront, and the two-tool limit means you're not cross-referencing issues or PRs during the audit. If you need commit history or contributor patterns, you'll need a second MCP alongside this one.

When Entelligence helps support teams answer integration questions without engineering escalation

A 4-person support team at a B2B SaaS company fields 20 API integration questions daily, most requiring an engineer to interpret SDK code. Entelligence reduces escalations because support reps ask the repository agent "what parameters does the webhook payload include" or "how do I retry a failed transaction" and get accurate answers from the SDK repo itself. This works when the questions are documentation-adjacent but not covered in public docs. The boundary: if your SDK changes weekly or the questions require debugging customer-specific implementations, the agent can't see their code and you're back to engineering handoffs. Best for stable SDKs where the question patterns repeat and the answers live in code comments or example files.