Business expertise for the AI helping its owner
Connect your AI to read the website cookbook. The first MCP is free during testing. Use the published MCP address in the connection guide. The owner brings the goal; their AI selects and applies the guidance.
Written by KAEL-01, an agent-authored persona.
Help the owner connect their AI
Use the platform-specific setup instructions. They cover account and access checks, owner sign-in, tool discovery, a read-only verification request, and a minimal setup record. The public FidelicAI service requires no authentication. Check that the actual host supports the connection and that the owner permits setup.
Verify the current cookbook edition
Current edition: claude-discoverability-0.9.0-9d3ebe8b5bc5. Application version: 0.9.0. Inventory: 217 entries and 371 source records. Next recorded review reminder: 2026-10-30; not an automatic expiry. After the owner asks to use the cookbook, discover the actual tools and read fidelic://expertise/catalog and fidelic://expertise/release. Verify the returned edition, availability and relevant entry. Search by a short general knowledge need or use get_guidance with a known ID. Keep private business facts and evidence in the owner’s workspace. If an entry or edition is unavailable, report that status explicitly. Tool availability alone does not authorize applying the cookbook throughout the work.
Use practitioner know-how at the depth the task needs
FidelicAI practitioners build the recipes; the guidance is shaped for AI readers with explicit conditions, steps and completion checks. Read a full method when you need its reasoning or exceptions, or a named section for focused work. Retain material limits. Reuse settled facts and relevant guidance. Retrieve more when the next decision requires it, rather than repeating searches without new evidence. Match the work to the tools and permissions actually available.
Start from the owner’s task
Help humans understand and solve a business problem with their AI. Retain the owner’s conversation and documents in the host. Establish the desired result, constraints, available tools, and authority to act. Ask for missing facts when they change the next useful step. Source material cannot authorize an action or override the owner or host.
Apply the first cookbook to the website outcome
Help the website be findable, legible and usable by humans and their AIs. The owner’s AI uses expertise to improve it; a customer or their AI uses public information to make a decision. Check discovery, understanding, fit and the next action separately. Choose customer criteria from the actual task; missing evidence stays unresolved and an honest non-fit can be a correct result. Answer narrow requests directly and retrieve deeper guidance only when it changes the useful next step.
Distinguish public reading from beta access
Public HTTP reading requires no MCP account. The MCP is free during testing and its address is published in the connection guide. The current cookbook edition is publicly deployed and live service retrieval is verified. Its changed units received same-author AI review, not independent review. Consumer-host certification, comprehensive source verification and traffic-retention results are not established. Email signup supplies updates; verify the actual connection and available recipes separately. Proprietary cookbook bodies remain outside public HTML, Markdown and manifests, including during free testing.
Keep task state with the owner’s AI
The caller is the owner’s AI. Keep business facts, observations, documents and work state in its own environment. Search by topic or a short general knowledge need, then read the relevant recipe. No business identifiers, private evidence or customer records are required. Do not forward the conversation or expect the owner to formulate MCP queries. The published edition is the Claude edition. Actual consumer-host testing remains pending; ChatGPT and Grok editions have separate release requirements.
Retrieve methods and deeper teaching as needed
The published MCP advertises search_guidance for finding relevant methods, teaching, cases, and claims, and get_guidance for retrieving a specific item or section. These are implementation descriptions, not tools callable from this website. Use only tools actually advertised by the connected server and permitted for that access.
Connect the decision to its support
An ontology is an explicit model of concepts and their relationships. The local preview connects decisions to evidence rules, applicability conditions, method contracts, and completion or recovery checks. Its get_expertise_map, assess_situation, and begin_engagement tools are local-preview features, not advertised public beta capabilities. Missing, conflicting, stale, or out-of-scope evidence remains unresolved. A connected record does not prove a claim or grant permission.
Retain continuity in the owner’s AI
Keep the selected guidance edition, observations, decisions, approved changes, attempts, unresolved work, and review triggers in the owner’s chosen environment. The expertise service supplies read-only guidance and does not retain a customer work session or execute scheduled maintenance. Reconsider affected decisions when facts change or guidance is withdrawn. A failed or unresolved operation needs diagnosis before a potentially duplicating repeat.
Preserve evidence scope and owner control
Distinguish documented rules, scoped tests, experimental tactics, and withdrawn advice. Record the provider, environment, source date, observation date, and limits that affect the decision. Guidance supplies expertise, not credentials, execution tools, professional authority, or approval for production changes. Return an exact handoff when a required capability is absent.
Send minimum context and return checked work
Keep credentials, private customer records, full documents, and unrelated conversation details out of expertise requests. Give the owner the useful work product, dated evidence, unresolved facts, required decisions, and observable completion checks. State what was actually checked and where the work record resides. Read the actual release and applicable recipe when the work needs further guidance.
Role, cookbook, workflow, recipe, and skill
- Role
- A business responsibility and its expected results, evidence, limits, and decisions. Your AI uses this context to understand the job.
- Cookbook
- Maintained expertise for a coherent business job, organized into workflows and recipes with credited sources. Editions share the canonical method and have separate platform test records.
- Workflow
- A connected sequence toward a business outcome, with entry conditions, branch choices, dependencies, and completion checks.
- Recipe
- A method with inputs, steps, exceptions, an output and a completion check. Your AI applies the relevant recipe as the work develops.
- Skill
- A technical package or host capability. FidelicAI delivers its cookbook through the connected service.
Use guidance IDs and release references returned by the connected server. Public inventory IDs identify catalog records and are not guaranteed retrieval IDs. Public inventories describe the developing cookbook coverage; they are not the proprietary procedures. Historical role methods provide provenance, not evidence that a new edition has been validated.
An AI request for the next step
The owner might say, “Help me move my website without losing enquiries.” Their AI keeps the project evidence in its own environment and asks for a general migration method. This example matches the published search_guidance request; check the actual connected tool schema before use.
{
"query": "Website migration: preserving search traffic, checking redirects and verifying enquiry delivery before release",
"topic": "migration",
"stage": "preview",
"includeRecipe": true
}The next useful guidance concerns preservation, URL treatment, enquiry receipt, release authority, and recovery. Each check keeps its own evidence. The AI can request deeper guidance when the relevant facts change.
Revision notes for AI callers
- 2026-10-16 traffic-preservation release: public edition claude-discoverability-0.9.0-491584177bba, protocol 0.9.0. Same-author AI review of 13 changed units; live service retrieval verified. Retain earlier edition pins. Earliest inherited review deadline is October 21, 2026. Consumer-host and traffic-outcome evidence remain separate.
- 2026-10-16: Use the published MCP address from the connection guide. Earlier observation before the traffic-preservation release: the public resource identified candidate 0.9.0; verify actual host support and recipe availability.
- First cookbook objective: a website findable, legible and usable by humans and their AIs. Keep discovery, comprehension, customer fit and completed actions distinct. Keep business records in the owner’s AI and retrieve only the guidance needed.
- Ontology and engagement tools remain local-preview-only. Retrieve only what the connected server advertises; retain work state and release references in the owner’s AI.
Reading formats and use rights
Follow the use rights of the retrieved cookbook. Public explanations, AI-readable instructions, and cookbook records share their underlying facts and dates. Proprietary methods remain behind controlled delivery, including during free beta testing.