CSV Contract Preflight MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| APIFY_TOKEN | No | Your existing eligible Apify buyer account token, configured via environment/configuration. Required for buyer-profile draft operations and purchases; without it, initialization and free tool discovery still work. | |
| CSV_CONTRACT_MCP_STATE | No | Optional path selecting the private operations state database (operations.sqlite3). Its parent must be owned by your OS user with mode 0700, and the database with mode 0600. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Agent-assisted drafting uses YOUR model: prepare, generate proposal, review, explicitly accept digest, then export or separately authorize a paid validation. This server does not run an owner-funded LLM. Three free explicit presets and a separate deterministic summary tool are available. Legacy reports, tool schemas, draft acceptance and purchase confirmation are unchanged.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| start_validationA | PURCHASE: explicitly start one bounded CSV report using YOUR Apify account. Completed pass/findings reports costUSD0.01. Never retry an uncertain start with another ID. Inputs stored by Apify. No repair/import permission. |
| get_validationB | Read only the registered original run and bounded matching dataset/OUTPUT; persist local result. Unknown start stays HOLD. No run creation, retry, input repair, creator-revenue or import claim. |
| prepare_contract_draftA | FREE LOCAL PREPARATION ONLY, not model generation. Return a bounded generationRequest for YOUR agent model using requirements and literal headers. Then submit its actual proposal with review_contract_draft. No purchase or automatic acceptance. |
| review_contract_draftB | Review YOUR model proposal JSON against the public contract profile and exact headers; store a pending preview. Does not accept it, generate with a server model, buy a report or repair data. Rejected proposals stay rejected; do not silently weaken requirements. |
| get_contract_draftB | Read the original local draft operation under this configured buyer credential profile. No model request or purchase. |
| accept_contract_draftA | EXPLICIT CALLER ACCEPTANCE of the exact preview/requirements/assumptions digest after semantic review. Questions block acceptance. This never buys, imports or certifies source truth. |
| export_accepted_contractB | Export canonical contractJson ONLY after explicit acceptance in this buyer state. No purchase/import. You may use it with the existing native Apify Actor through your own authorized account. |
| start_validation_from_draftB | PURCHASE one native Apify validation using an explicitly accepted local draft and YOUR Apify account. Separate confirmPurchase=true is mandatory. Completed pass/findings reports costUSD0.01. Preserve original state and never retry unknown starts. |
| list_contract_templatesA | FREE local catalog of three explicit supported-profile presets. No token, state, model call, acceptance, or purchase. |
| get_contract_templateB | FREE local explicit preset: exact headers/rules, canonical contract, proposal, synthetic pass/findings samples and limitations. Inspect its rules. Selecting it does not accept a draft or buy a report. Reuse prepare/review/accept for that workflow; confirmPurchase remains separate. |
| get_validation_summaryA | Explain an EXISTING local validation operation, retrieving only its original known run if needed. Revalidate the completed report against original hashes; return deterministic counts and static correction hints. Incomplete/held/failed operations have summary:null. May update local result state; never creates or retries a paid run, alters the report, repairs data, or authorizes import. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 11 tools
The draft lifecycle (prepare/review/get/accept/export) and template list/get pairs are cleanly separated, and the descriptions clearly delimit paid vs free operations. Two near-duplicate pairs create mild ambiguity: start_validation vs start_validation_from_draft, and get_validation vs get_validation_summary, where an agent must read carefully to pick the right read path.
All tools use snake_case verb_noun form with a small, stable verb set (start, get, prepare, review, accept, export, list). Multi-word objects like contract_draft and validation are applied consistently across the lifecycle.
Eleven tools is well within a reasonable range for a two-phase workflow (draft creation then paid validation). There is some redundancy in maintaining parallel start_validation/start_validation_from_draft and get_validation/get_validation_summary paths, but each earns a distinct role.
CRUD-ish coverage exists for drafts (prepare, review, get, accept, export) and validation (start, get, summary, start-from-draft), but there is no way to list existing drafts or validations, so an agent must already know IDs. There is also no cancel/abort or delete operation for drafts or runs, leaving cleanup and discovery as dead ends.