"How to run code using Claude Desktop" matching MCP tools:

• setup_instructions

  Coal — Payments for AI agents

  Print step-by-step instructions for using Coal MCP from Claude / Cursor / any MCP client. Run this FIRST if you are unsure how to authenticate or which credentials to provide.

  Connector

• map_loinc_to_snomed

  Medical Terminologies MCP

  This tool looks up a LOINC code in NLM Clinical Tables and returns guidance on where to obtain a LOINC → SNOMED CT mapping. It does not perform the mapping. Direct LOINC → SNOMED CT mappings are not freely available via API. UMLS Metathesaurus contains the relationships but requires an individual UMLS Terminology Services license; the LOINC SNOMED CT Expression Association is published by Regenstrief Institute as part of the LOINC release and requires authenticated download from loinc.org under the LOINC license. For programmatic LOINC → SNOMED mapping, use UMLS or the LOINC Expression Association files. For interactive lookup, use the SNOMED CT browser available to your organization or the Regenstrief RELMA desktop tool. Provide a LOINC code like "2339-0" (Glucose) or "718-7" (Hemoglobin).

  Connector

• upload_media

  SendIt

  FOR CLAUDE DESKTOP ONLY (with filesystem access). For Claude.ai/web: Use create_upload_session instead - it provides a browser upload link. Upload local media to cloud storage, returning a public HTTPS URL. WHEN TO USE: • Instagram, LinkedIn, Threads, X: REQUIRED for local files before calling publish_content • TikTok: NOT NEEDED - pass local path directly to publish_content SUPPORTED FORMATS: • Images: jpg, png, gif, webp (max 10MB) • Videos: mp4, mov, webm (max 100MB) Returns { url: 'https://...' } for use in publish_content mediaUrl parameter.

  Connector

• tfgenerate

  InsideOut (Riley)

  WORKFLOW: Step 3 of 4 - Generate Terraform files from completed design Generate Terraform files from an InsideOut session that has completed infrastructure design. ⚠️ PREREQUISITE: Only call this AFTER convoreply returns with `terraform_ready=true` in the response metadata. DO NOT call this while convoreply is still running or before terraform_ready is confirmed! If you get 'session has not reached terraform-ready state', wait for convoreply to complete first. 🎯 USE THIS TOOL WHEN: convoreply has returned with terraform_ready=true, OR the user asks to 'see the terraforms', 'generate terraform', 'show me the code', etc. **DEFAULT RESPONSE**: Returns summary table + download URL (keeps code out of LLM context). **FALLBACK**: Set `include_code: true` to get full code inline if curl/unzip fails. **CRITICAL WORKFLOW** (default mode): 1. Call this tool to get file summary and download URL 2. ASK the user: 'Where would you like me to save the Terraform files? Default: ./insideout-infra/' 3. WAIT for user confirmation before running the download command 4. Run the curl/unzip command with the user's chosen directory 5. If curl/unzip FAILS (sandbox, security, platform issues), retry with `include_code: true` **AFTER GENERATION**: Ask user if they want to review the files and then deploy with tfdeploy REQUIRES: session_id from convoopen response (format: sess_v2_...). OPTIONAL: include_code (boolean) - set true to return full code inline as fallback. 💡 TIP: Examine workflow.usage prompt for more context on how to properly use these tools.

  Connector

• get_integration_guide

  TestMyVibes

  Returns the canonical guide for using TMV from a coding-agent context. Covers the fix-test-retest loop, how to write a good test prompt, how to read the actionTrail / consoleErrors / failedRequests outputs, and common gotchas. Call this first if you're a new agent on a project — it'll save you a debug session. The same content is served at https://testmyvibes.com/docs/coding-agents.

  Connector

• architect.certify

  AI Design Blueprint Doctrine

  Pro/Teams — second-pass adversarial certification of an architect.validate run that scored production_ready (A or B first-pass tier). ON CLIENT TIMEOUT — DO NOT RETRY THIS TOOL. **RECOVERY FIRST**: the run_id is emitted in the FIRST notifications/progress event at t=0s (BEFORE the LLM call begins). Capture it. On timeout, call `me.validation_history(run_id='<that-id>')` to fetch the persisted cert verdict; the server-side run completes independently within a 20-minute budget. This is the canonical recovery path. Use it before considering any retry. Long-running LLM call (60-180s typical; exceeds Claude Code's ~60s idle budget); MCP clients commonly close the call before the server returns. Retrying re-runs the LLM call AND burns one of your 3 cert retry-budget attempts. Mints the certified production_ready badge when both reviewers sign off; caps the run to C/emerging when the second pass surfaces a missed production_blocker. MANDATORY DOCTRINE RULE (load-bearing): the badge certifies the EXACT code that produced the validate run_id, NOT 'this codebase' in general. If you modify, fix, or iterate the code between architect.validate and architect.certify — even a single character — cert rejects with code_fingerprint_mismatch. Fixing the code voids the run. The recovery path is always: edit code → architect.validate → fresh run_id → architect.certify on the fresh run. Do NOT cert from a stale run_id after iteration; ask the user to re-validate first. WHEN TO CALL: only after architect.validate returned tier=production_ready AND the user wants the certified badge AND the code has not been touched since the validate run. NOT for tier=draft/emerging/not_applicable runs (typed rejections fire — see below). NOT idempotent across attempts: each call is one of the 3 attempts in the retry budget. BEHAVIOR: atomic one-shot single LLM call, ~60-180s server-side at high reasoning effort (small payloads finish faster; observed p99 ~250s; server-side budget is 20 min, ~5× observed max). Exceeds typical MCP-client tool-call idle budget (~60s in Claude Code), so the FIRST notifications/progress event fires at t=0 carrying the run_id. The run is atomic by contract — no in_progress lifecycle, no cancellation, no resume. Updates the persisted run's result_json (public review URL + me.validation_history(run_id=...) reflect the cert outcome). ELIGIBILITY GATE (typed rejection enum on failure): caller must own the run, tier=production_ready, less than 24h old, not already certified, within cert retry budget (max 3 attempts), no other cert call in flight for the same run_id, code fingerprint must match the validated code, AND the submitted payload must be cert-payload-complete (see Payload Completeness below — cert rejects pre-LLM with `payload_incomplete` when an imported module's surface isn't visible in the validate payload that produced this run_id). Rejection reasons (typed Literal): auth_required, paid_plan_required, run_not_found, not_run_owner, not_eligible_tier, not_agentic_component (tier=not_applicable runs), already_certified, certification_age_exceeded, retry_budget_exhausted, code_fingerprint_mismatch, code_fingerprint_missing, code_not_on_file (caller omitted `code` argument AND the 24h cert-retry hold for this run has expired or was never written. Recovery: re-run architect.certify from the same MCP session that ran architect.validate, passing the code explicitly — the server never persists code by design), payload_incomplete (submitted/validated payload imports modules whose contents aren't visible — cert refuses pre-LLM to prevent a false-precision downgrade. Recovery: re-validate with verbatim public-surface stubs for every imported module, then re-cert on the fresh run_id. Empirically validated: PR #157 iter8/iter9 cert rejections were exactly this class — code on disk was correct, the submitted payload merely omitted module visibility), cert_consensus_score_below_threshold (consensus_median<75 — consensus runs only), cert_consensus_unstable_blocker (any principle mode_stability<80% — consensus runs only), run_state_corrupt, cert_persistence_failed, cert_in_flight (a prior architect.certify call on this run_id is still running. Poll me.validation_history for the verdict; do not retry until it resolves). PAYLOAD COMPLETENESS (load-bearing for cert eligibility): the cert reviewer reads the EXACT payload that produced the validate run_id. Imported modules whose surface isn't present in the payload cause pre-LLM `payload_incomplete` refusal. Avoidance — when validating with intent to cert, bundle public-surface stubs for every imported module: `from sqlalchemy.exc import SQLAlchemyError` → include a stub class; `from app.db import models` → include a `class models:` namespace stub with the columns/methods you reference; module-level imports of `dataclass`, `Literal`, `json`, `datetime`, `timezone` MUST also be in the payload (cert correctly catches when they're omitted — code would NameError on import). 'Submit Like Production': the payload should be the code as it would actually run, not a compressed sketch. PRE-LLM REJECTION AUDIT TRAIL: when cert rejects before the LLM call (payload_incomplete, code_fingerprint_mismatch, etc.), `certification_attempts=[]` on the response — no attempt landed in the retry budget, no LLM hop occurred. The rejection envelope's `rejection_reason` + `guidance` are the actionable surface. (Audit-trail UI surfacing of pre-LLM rejections is tracked in the platform self-audit set as anomaly #5; out of scope for the cert tool itself.) INPUTS: re-send the SAME code that produced the run_id (the architect persists findings + recommendations, never code, by design — privacy-preserving). Server compares the submitted code's SHA-256 fingerprint to the stored fingerprint and rejects mismatches. Auth: Bearer <token>, Pro or Teams plan required. UK/EU data residency (Cloud Run europe-west2). Code processed transiently by OpenAI (no-training-on-API-data) and dropped; payloads JSON-escaped + delimited as inert untrusted data — prompt-injection inside code is ignored. If the cert call fails outright (provider error, persistence error), a fresh architect.certify is the recovery path; the eligibility gate enforces the 3-attempt retry budget. For long-running cert workflows the answer is to re-validate, not to make this tool stateful. OUTCOMES: certification_status ∈ {confirmed_production_ready (badge mints), downgraded_to_emerging (cert review surfaced a missed production_blocker, tier capped at C/emerging), unavailable_provider_error (LLM call failed, retry within budget)}. Cert findings + summary + attempt history surfaced on the persisted run for full inspectability.

  Connector

• discovery_signup_verify

  Disco

  Complete Disco signup using an email verification code. Call this after discovery_signup returns {"status": "verification_required"}. The user receives a 6-digit code by email — pass it here along with the same email address used in discovery_signup. Returns an API key on success. Args: email: Email address used in the discovery_signup call. code: 6-digit verification code from the email.

  Connector

• prepare_image_upload

  Avocado AI

  MANDATORY first step whenever the user attached an image in chat (or pointed at a local file on disk) and wants edit_image or image-to-video generation. Returns a signed PUT URL plus a file_id. After this tool: either (a) the inline upload widget will let the user drop the file and auto-continue (Claude.ai web), or (b) you run a curl PUT yourself if you have shell access (Claude Desktop / Claude Code) — the response text contains a ready-to-run curl command. Then call edit_image or generate_video with file_id=<returned id>. edit_image and generate_video do NOT accept base64 — calling them with raw image bytes WILL fail. This tool is the only working path for chat attachments. Set `purpose` to 'edit' or 'video' so the upload widget points the user at the right downstream tool.

  Connector

• architect.certify

  AI Design Blueprint

  Pro/Teams — second-pass adversarial certification of an architect.validate run that scored production_ready (A or B first-pass tier). ON CLIENT TIMEOUT — DO NOT RETRY THIS TOOL. **RECOVERY FIRST**: the run_id is emitted in the FIRST notifications/progress event at t=0s (BEFORE the LLM call begins). Capture it. On timeout, call `me.validation_history(run_id='<that-id>')` to fetch the persisted cert verdict; the server-side run completes independently within a 20-minute budget. This is the canonical recovery path. Use it before considering any retry. Long-running LLM call (60-180s typical; exceeds Claude Code's ~60s idle budget); MCP clients commonly close the call before the server returns. Retrying re-runs the LLM call AND burns one of your 3 cert retry-budget attempts. Mints the certified production_ready badge when both reviewers sign off; caps the run to C/emerging when the second pass surfaces a missed production_blocker. MANDATORY DOCTRINE RULE (load-bearing): the badge certifies the EXACT code that produced the validate run_id, NOT 'this codebase' in general. If you modify, fix, or iterate the code between architect.validate and architect.certify — even a single character — cert rejects with code_fingerprint_mismatch. Fixing the code voids the run. The recovery path is always: edit code → architect.validate → fresh run_id → architect.certify on the fresh run. Do NOT cert from a stale run_id after iteration; ask the user to re-validate first. WHEN TO CALL: only after architect.validate returned tier=production_ready AND the user wants the certified badge AND the code has not been touched since the validate run. NOT for tier=draft/emerging/not_applicable runs (typed rejections fire — see below). NOT idempotent across attempts: each call is one of the 3 attempts in the retry budget. BEHAVIOR: atomic one-shot single LLM call, ~60-180s server-side at high reasoning effort (small payloads finish faster; observed p99 ~250s; server-side budget is 20 min, ~5× observed max). Exceeds typical MCP-client tool-call idle budget (~60s in Claude Code), so the FIRST notifications/progress event fires at t=0 carrying the run_id. The run is atomic by contract — no in_progress lifecycle, no cancellation, no resume. Updates the persisted run's result_json (public review URL + me.validation_history(run_id=...) reflect the cert outcome). ELIGIBILITY GATE (typed rejection enum on failure): caller must own the run, tier=production_ready, less than 24h old, not already certified, within cert retry budget (max 3 attempts), no other cert call in flight for the same run_id, code fingerprint must match the validated code, AND the submitted payload must be cert-payload-complete (see Payload Completeness below — cert rejects pre-LLM with `payload_incomplete` when an imported module's surface isn't visible in the validate payload that produced this run_id). Rejection reasons (typed Literal): auth_required, paid_plan_required, run_not_found, not_run_owner, not_eligible_tier, not_agentic_component (tier=not_applicable runs), already_certified, certification_age_exceeded, retry_budget_exhausted, code_fingerprint_mismatch, code_fingerprint_missing, code_not_on_file (caller omitted `code` argument AND the 24h cert-retry hold for this run has expired or was never written. Recovery: re-run architect.certify from the same MCP session that ran architect.validate, passing the code explicitly — the server never persists code by design), payload_incomplete (submitted/validated payload imports modules whose contents aren't visible — cert refuses pre-LLM to prevent a false-precision downgrade. Recovery: re-validate with verbatim public-surface stubs for every imported module, then re-cert on the fresh run_id. Empirically validated: PR #157 iter8/iter9 cert rejections were exactly this class — code on disk was correct, the submitted payload merely omitted module visibility), cert_consensus_score_below_threshold (consensus_median<75 — consensus runs only), cert_consensus_unstable_blocker (any principle mode_stability<80% — consensus runs only), run_state_corrupt, cert_persistence_failed, cert_in_flight (a prior architect.certify call on this run_id is still running. Poll me.validation_history for the verdict; do not retry until it resolves). PAYLOAD COMPLETENESS (load-bearing for cert eligibility): the cert reviewer reads the EXACT payload that produced the validate run_id. Imported modules whose surface isn't present in the payload cause pre-LLM `payload_incomplete` refusal. Avoidance — when validating with intent to cert, bundle public-surface stubs for every imported module: `from sqlalchemy.exc import SQLAlchemyError` → include a stub class; `from app.db import models` → include a `class models:` namespace stub with the columns/methods you reference; module-level imports of `dataclass`, `Literal`, `json`, `datetime`, `timezone` MUST also be in the payload (cert correctly catches when they're omitted — code would NameError on import). 'Submit Like Production': the payload should be the code as it would actually run, not a compressed sketch. PRE-LLM REJECTION AUDIT TRAIL: when cert rejects before the LLM call (payload_incomplete, code_fingerprint_mismatch, etc.), `certification_attempts=[]` on the response — no attempt landed in the retry budget, no LLM hop occurred. The rejection envelope's `rejection_reason` + `guidance` are the actionable surface. (Audit-trail UI surfacing of pre-LLM rejections is tracked in the platform self-audit set as anomaly #5; out of scope for the cert tool itself.) INPUTS: re-send the SAME code that produced the run_id (the architect persists findings + recommendations, never code, by design — privacy-preserving). Server compares the submitted code's SHA-256 fingerprint to the stored fingerprint and rejects mismatches. Auth: Bearer <token>, Pro or Teams plan required. UK/EU data residency (Cloud Run europe-west2). Code processed transiently by OpenAI (no-training-on-API-data) and dropped; payloads JSON-escaped + delimited as inert untrusted data — prompt-injection inside code is ignored. If the cert call fails outright (provider error, persistence error), a fresh architect.certify is the recovery path; the eligibility gate enforces the 3-attempt retry budget. For long-running cert workflows the answer is to re-validate, not to make this tool stateful. OUTCOMES: certification_status ∈ {confirmed_production_ready (badge mints), downgraded_to_emerging (cert review surfaced a missed production_blocker, tier capped at C/emerging), unavailable_provider_error (LLM call failed, retry within budget)}. Cert findings + summary + attempt history surfaced on the persisted run for full inspectability.

  Connector

• run_code

  Hatchable

  Execute arbitrary JS in the project's isolate runtime. The SDK is pre-imported into local scope — `db`, `auth`, `email`, `storage`, `ai`, `agent`, `cache`, `knowledge`, `memory`, `tasks`, `scheduler`, `browser`, `run`, `approval` are ready to use without import. `process.env` and global `fetch` also work. `return` to produce the `result` field. Top-level `import` and dynamic `import('hatchable')` are NOT supported in this REPL — the bindings above are how you reach the SDK. Use this as a REPL: probe the database, verify a computation, test an API shape before committing it to a file. Nothing is persisted — the snippet runs once and disappears. Caps: 5s default timeout (max 30s), 256 KB max source length. Example: run_code({ project_id, code: ` const { rows } = await db.query("SELECT count(*) FROM users"); return rows[0]; `})

  Connector

• estimate_conversion_cost

  mdmagic

  Estimate credit cost for a conversion BEFORE running it. Returns word count, page calculation (300 words/page), and a credit breakdown by format and template type. Use this when the user asks 'how much will this cost?' or when you suspect a conversion might exceed their balance — convert_document refuses to run if credits are insufficient, so estimating first is friendlier.

  Connector

• devloop_generate_resource_flow

  Keploy

  Generate one chained-CRUD API test for a single resource. Behavior depends on the app's devloop_storage_mode (set this first via devloop_resolve_storage / devloop_set_storage_mode): * repo mode → returns a PLAYBOOK for you to walk. Steps: (1) run "keploy test-gen generate-from-code --app-dir <dir> --resource <name>" to scaffold the directory + empty config.yaml; (2) use your Write tool to author keploy/api-tests/<resource>/test.yaml using the schema returned by devloop_detect_app; (3) run "keploy test-gen run --test-dir keploy/api-tests --suite <Name>_CRUD --base-url <url> --ci" to verify the test parses and passes; (4) call devloop_mutation_demo next (auto, per the DEVLOOP instructions). * cloud mode → returns guidance to call the existing create_test_suite tool instead. The repo-mode playbook is NOT used in cloud mode. ARGUMENTS — you should already have these from your devloop_detect_app call: * app_id, resource, app_dir, base_url, framework, handler_files. If any are missing, call devloop_detect_app again. The tool does NOT generate the YAML body itself — you do, using the schema from devloop_detect_app's detection_playbook. This is intentional: ATG quality depends on the AI seeing the actual handler implementations (which it can read via its own tools) far better than a server-side generator could. Aim for ≤ 30 lines per test.yaml, idempotent mutating steps, chained extract/{{var}} flow.

  Connector

• explore_design_conditions

  epwforge-mcp

  Interactive single-site design-conditions explorer. Returns full ASHRAE design conditions + diurnal chart for the requested scenario. In MCP Apps-capable hosts (Claude Desktop, ChatGPT, VS Code, Goose), the response renders as a widget with sliders for SSP / year / percentile / UHI — dragging a slider re-calls this tool live. Use when a user wants to interactively tune a single site. For multi-site comparison, use analyze_weather(urls=[...]) instead. Defaults to present-day TMY (no morph) — pass ssp+year for future scenarios. P75 default percentile is design-realistic; P50 underestimates the tail. No auth required.

  Connector

• retry_run

  Hydrata - ANUGA Flood Simulation

  Retry a failed simulation run. Resets an errored run back to 'created' status and triggers a new package build. The same run ID is reused. Only valid when status is 'error'. Returns 409 for any other state.

  Connector

• discover_commerce_tools

  AXIS Toolbox — Agentic Commerce Codebase Intelligence

  Discover AXIS install metadata, pricing, and shareable manifests for commerce-capable agents. Free, no auth, and no mutation beyond read access. Example: call before wiring AXIS into Claude Desktop, Cursor, or VS Code. Use this when you need onboarding and ecosystem setup details. Use search_and_discover_tools instead for keyword routing or discover_agentic_purchasing_needs for purchasing-task triage.

  Connector

• edit_image

  Avocado AI

  Modify an existing image. REQUIRED input: exactly one of file_id OR image_url. base64 is NOT accepted — do not try to pass image bytes as a tool argument, the call will be rejected. For chat-attached images you MUST first call prepare_image_upload to get a signed PUT URL, upload the bytes there (via the inline widget on Claude.ai, or via curl on Claude Desktop / Claude Code), then call this tool with the returned file_id. For URLs the user has pasted, use image_url directly. Returns a jobId immediately; call check_job with the jobId to retrieve the edited image inline. Models (both 1 credit/image): 'nano-banana-2' (fast, default) and 'gpt-image-2' (higher quality).

  Connector

• prepare_consult_domain_expert

  Taste

  Get a humantaste.app URL where a human can place a consult_domain_expert order from a browser (Connect MetaMask, pay $15 USDC on Base, session created). Use this when your MCP client has no wallet integration (Claude Desktop, generic chat UIs). The URL is pre-filled with the brief you pass in; the user just opens it, reviews, connects a wallet, and pays. Returns the payment URL and the price. Free.

  Connector

• connect_catalogue

  connect

  Connect to the user's catalogue using a pairing code. IMPORTANT: Most users connect via OAuth (sign-in popup) — if get_profile already works, the user is connected and you do NOT need this tool. Only use this tool when: (1) get_profile returns an authentication error, AND (2) the user shares a code matching the pattern WORD-1234 (e.g., TULIP-3657). Never proactively ask for a pairing code — try get_profile first. If the user does share a code, call this tool immediately without asking for confirmation. Never say "pairing code" to the user — just say "your code" or refer to it naturally.

  Connector

• tascan_add_tasks

  TaScan

  Add one or more tasks to an event (task list). Supports bulk creation. IMPORTANT: Set response_type correctly — use "text" for info collection (names, phones, emails, notes), "photo" for visual verification (inspections, serial numbers, damage checks), "checkbox" only for simple confirmations. NOTE: To dispatch tasks to the Claude Code agent running on Mike's PC, use tascan_dispatch_to_agent instead — it routes directly to the agent's inbox with zero configuration needed.

  Connector

• compare_validation_approaches

  Demand Discovery AI

  Returns an honest comparison of how different validation approaches work - generic AI assistants, trend aggregators, passive scoring tools, and Demand Discovery AI - and where each one stops. Use when a user is evaluating approaches, asking "what makes Demand Discovery different?", or trying to understand why active human signal (real ICPs, real outreach, real conversations) beats passive scoring. Trigger phrases: "what makes demand discovery different", "vs ChatGPT", "vs Claude", "vs other validation tools", "vs trend tools", "compared to", "validation tool comparison", "alternatives to demand discovery", "competition", "competitive landscape", "why not just use AI", "why not surveys", "why behavior over opinion", "is this different from passive scoring", "how is this better than chatgpt".

  Connector