# Agent Interface

> We make existing software operable by AI agents. Category: AX (Agent Experience).
> Your MCP server works in the demo. Then an agent tries real work and stops at
> step three. We build the layer that gets it to the end.

## What we measure

Agent Coverage — the percentage of a product's real capabilities an agent can
complete unaided, end to end. Measurable from the outside. Typical unaudited
score: under 10%.

## Free entry offer (lead with this)

We run a real agent against your product and send back, at no cost:

1. your Agent Coverage score — what share an agent can finish on its own
2. exactly where it broke, step by step
3. a short mock of what your agent layer looks like once it's fixed

No call required to receive it. Reply and we run it. It takes about a day.
This is the way in — the paid work below only comes after you've seen it.

## Services

- Agent Readiness Audit — $2,500, 2 weeks. Scored report + prioritized fix list. Credited in full against a build.
- MCP Rescue — $10,000, 3 weeks. Fix a shipped-but-shallow server: auth, ambiguity, gates, error design, surface expansion.
- Agent-Ready in 30 Days — $15,000, 30 days. Full layer: tool surface, skills/docs, auth, discovery, /agents page, evals, measured coverage lift.
- Ongoing reliability retainer — from $2,500/mo. Regression evals on model releases, coverage reporting, discovery maintenance.

## Failure modes we audit for

WRONG_GRANULARITY   tools mapped to endpoints, not jobs
CONTEXT_BLOAT       tool descriptions re-sent every turn, eating the window
DOC_VOICE           descriptions written as documentation, not instructions
OPAQUE_ERRORS       errors that carry no recovery instruction
AMBIGUOUS_NAMING    near-identical tool names the agent cannot distinguish
RAW_OUTPUT          unfiltered dumps the agent must reason over
NO_SEQUENCE         no encoded order of operations
NOT_IDEMPOTENT      retries and double-calls corrupt state
NO_GATES            nothing stops an agent publishing, deleting, or spending
NEVER_EVALED        tested for 200s, never for goal completion

## Proof

VoxHire.ai (recruiting platform) — Agent Coverage 4% before, 71% after.
Golden path: whoami -> list_jobs -> list_applications -> get_token_usage ->
create_agent_template -> start_agent -> get_evaluations -> bulk_tag_applications

## Machine surfaces

/llms.txt                      this file
/llms-full.txt                 full corpus, one fetch
/.well-known/mcp.json          server card
/.well-known/agent-card.json   A2A card
/.well-known/ai-catalog.json   resource discovery
/agents/docs/tools.md          tool reference
/agents/docs/skills.md         how to evaluate agent-readiness
/agents/docs/vocabulary.md     terms we use
/agents/docs/examples.md       worked traces

## Contact

Humans and agents welcome. Phone: +91 79896 81404