Related Pages

Related topics: MCP Tools Catalog & Data Sources, Installation, Operations, Security & Extensibility

Overview, Architecture & Payment Flow

Purpose and Scope

blockrun-mcp is a Model Context Protocol (MCP) server that lets AI agents — primarily Codex and Claude — pay per call for live data and 30+ frontier AI models without managing API keys. It is published as @blockrun/mcp (current version: 0.21.1) and is shipped as an ESM binary (bin: blockrun-mcp) that runs on Node >= 18 Source: [package.json].

The server's stated surface is "web search, deep research, prediction markets, crypto data, X/Twitter intelligence" — all settled through x402 micropayments in USDC rather than subscriptions or rate-limited keys Source: [package.json]. Skills distributed alongside the server (skills/search, skills/surf, skills/exa-research, skills/image-prompting) describe how agents should compose tool calls for specific research workflows.

A community request to extend x402 settlement beyond Base (to Algorand, Hedera, Stellar, VOI, Tempo alongside Solana) is tracked in issue #13 — confirming that x402 USDC on Base is the canonical settlement rail today and Solana is the second supported chain.

System Architecture

The project follows a thin-MCP, fat-SDK split. blockrun-mcp is a small transport/proxy layer: it registers MCP tools, validates arguments, manages a local USDC wallet, enforces per-session budgets, and forwards the request through @blockrun/llm (^2.11.0) which actually speaks the x402 protocol Source: [package.json; AGENTS.md]. Upstream surfaces include xAI Grok Live Search, Exa neural search, the 84-endpoint Surf crypto/prediction-market graph, and image-generation models.

The repo structure makes this split explicit Source: [AGENTS.md]:

src/
├── index.ts          # MCP server entry point
├── mcp-handler.ts    # MCP protocol handler
├── tools/            # MCP tool implementations
├── types.ts          # Type definitions
└── utils/            # Shared utilities (wallet, budget, errors)

Tools are path-based passthroughs rather than bespoke schemas. As of v0.14.1 the blockrun_exa tool takes { path, body } and forwards to /v1/exa/<path> Source: [skills/exa-research/SKILL.md]. The same pattern shows up in blockrun_surf (84 path-based endpoints under /v1/surf/...) Source: [skills/surf/SKILL.md] and blockrun_search for Grok Live Search Source: [src/tools/search.ts]. This keeps the MCP schema minimal and lets new endpoints ship server-side without a client upgrade.

sequenceDiagram
    participant Agent as AI Agent (Codex/Claude)
    participant MCP as blockrun-mcp
    participant Wallet as utils/wallet.ts
    participant Budget as utils/budget.ts
    participant SDK as @blockrun/llm
    participant Upstream as Upstream (Grok / Surf / Exa / Images)
    participant Chain as Base / Solana (USDC)

    Agent->>MCP: tool call (path, body)
    MCP->>Wallet: getClient()
    MCP->>Budget: checkBudget(estimate)
    Budget-->>MCP: allowed / denied
    MCP->>SDK: requestWithPaymentRaw(endpoint, body)
    SDK->>Upstream: HTTP request
    Upstream-->>SDK: 402 Payment Required
    SDK->>Chain: pay USDC (x402)
    Chain-->>SDK: tx receipt
    SDK->>Upstream: retry with payment header
    Upstream-->>SDK: response
    SDK-->>MCP: result
    MCP->>Budget: recordSpending(actual)
    MCP-->>Agent: tool result

x402 Payment Flow

Every paid call follows the same five-step pattern visible in the sequence diagram above. Three pieces of the codebase make this work.

Wallet bootstrapping. utils/wallet.ts exposes getClient() which the tools use to obtain a payment-capable client on demand. On first launch the server generates a non-custodial USDC wallet (Base or Solana per the agent's choice), prints a deposit address, and renders a QR code via the qrcode package for easy funding Source: [package.json]. Surf settlement always lands in Surf's Base treasury even when the agent's wallet is on Solana Source: [skills/surf/SKILL.md].

Budget gating. Before any network call, utils/budget.ts::checkBudget() is consulted with an upfront cost estimate; after a successful response, recordSpending() debits the actual cost. The search tool illustrates the convention: estimateSearchCost(body) multiplies the requested max_results by a per-source price constant and clamps to the upstream 1–50 window Source: [src/tools/search.ts].

Per-call pricing. Costs are published next to the skill that documents them. Grok Live Search is $0.025 per returned source, charged by max_results rather than by sources actually returned, so max_results: 20 costs $0.50 even when the summary cites one link Source: [skills/search/SKILL.md]. Surf's chat/completions is a flat $0.02 in v1 (per-token billing deferred to Phase 2) Source: [skills/surf/SKILL.md]. Image generation ranges from ~$0.01 (google/nano-banana) to ~$0.06 (black-forest/flux-1.1-pro), with openai/gpt-image-2 at ~$0.04 and the only model with valid landscape/portrait/square sizes 1024x1024, 1536x1024, 1024x1536 Source: [skills/image-prompting/SKILL.md].

A community feature request (#14) proposes a free preflight trust check before any model routing — a BLOCK/CAUTION/PROCEED verdict on the destination payTo wallet, which would slot into the checkBudget → requestWithPaymentRaw boundary above as an additional zero-cost gate.

Tool Surface

Missing required parameters surface as a 400 with no charge; only successful payments produce a debit, so failed routing costs nothing. Surf's onchain/sql endpoint is explicitly noted as unrestricted on the server side — callers should add their own LIMIT or risk paying for megabytes of JSON Source: [skills/surf/SKILL.md]. A Security Advisory channel has been requested in issue #11 but no SECURITY.md is currently shipped.

See Also

• Skills reference: skills/search, skills/surf, skills/exa-research, skills/image-prompting — each documents the path table, pricing, and worked examples for one tool family.
• SDK: @blockrun/llm on npm — the x402 payment layer that blockrun-mcp wraps.
• Project home: <https://github.com/blockrunai/blockrun-mcp>
• Companion Python SDK used in skill examples: pip install blockrun-llm.

Related Pages

Related topics: Overview, Architecture & Payment Flow, Skills, Endpoint Catalogs & Prompting, Installation, Operations, Security & Extensibility

MCP Tools Catalog & Data Sources

Overview

@blockrun/mcp (v0.21.1) is an MCP (Model Context Protocol) server that exposes a paid catalog of generative AI, search, and crypto data tools to LLM agents. Every tool call settles through x402 USDC micropayments on Base, so there are no API keys to provision and no traditional rate limits. The server registers all tools through a single handler that wires each tool to a shared BudgetState and a payment-capable client from @blockrun/llm (Source: src/mcp-handler.ts:9-32). Source: package.json:3-22

The tool surface is organized around two complementary ideas documented in the project: MCP tools (typed, parameter-validated wrappers around partner APIs that may need local logic such as budget checks or multi-step polling) and skills (Markdown prompt-only references that document a large endpoint catalog without expanding the JSON schema the model has to parse). The contributing guide states this rule explicitly: "Add an MCP tool when the API needs typed parameter validation, has complex client logic, or is core to BlockRun value-prop; add a skill for lightweight passthrough with a large catalog." Source: CONTRIBUTING.md:17-28

Tool Catalog

The MCP handler wires 18 tool registrations, each implementing the same shape — registerXTool(server, budget) (Source: src/mcp-handler.ts:15-32). The catalog groups naturally into four functional families:

Source: src/mcp-handler.ts:9-32. The Surf tool alone is documented as exposing an 84-endpoint catalog covering token metrics, project DeFi rankings, prediction markets (Polymarket + Kalshi), wallet profiles, social/Twitter intelligence, and on-chain SQL via ClickHouse (Source: skills/surf/SKILL.md:1-50).

Architecture: Path-Based Passthrough

Tools with a large partner catalog (Surf, Exa) are deliberately implemented as path-based passthroughs so the MCP tool description stays small and new endpoints can be added without an MCP release. The tool description is intentionally minimal; the catalog lives in a companion skill the model reads on demand.

flowchart LR
    Agent[LLM Agent] -->|tools/call| MCP[MCP Server<br/>src/mcp-handler.ts]
    MCP -->|checkBudget| B[BudgetState]
    MCP -->|getClient| W[Wallet / x402]
    MCP -->|path + body| Tool[blockrun_surf / blockrun_exa]
    Tool -->|requestWithPaymentRaw| SDK[@blockrun/llm]
    SDK -->|x402 USDC| Partner[Partner API<br/>asksurf.ai / Exa / Grok]
    Partner -->|data| Tool
    Tool -->|recordSpending| B
    Tool -->|result| Agent

Source: src/tools/surf.ts:1-30, Source: src/tools/exa.ts:1-20. Two patterns repeat inside every tool:

1. Local budget pre-check — checkBudget(budget, estimateCost) is called before the upstream request. Surf uses an explicit tier table (T1 $0.01, T2 $0.02, T3 $0.02, T4 $0.03) keyed by endpoint path; Exa computes a cost from urls.length for contents and a flat $0.01 for other actions. This is a local guard only; the upstream settlement is authoritative (Source: src/tools/surf.ts:24-30, src/tools/exa.ts:14-21).
2. Method inference — if the caller supplies body, the tool POSTs; otherwise it GETs with params. Surf forwards to the partner treasury using BlockRun's server-held API key, while Exa settles through the standard x402 flow.

Skills, Cost, and Community Context

Skills are plain Markdown (skills/<name>/SKILL.md) loaded by the agent at runtime. They carry the *catalog*, *prompt frameworks*, and *gotchas* the model needs. For example, the image skill prescribes a 5-section prompt framework (SCENE / SUBJECT / DETAILS / USE CASE / CONSTRAINTS), documents the three valid GPT Image 2 sizes (1024x1024, 1536x1024, 1024x1536), and lists pricing tiers per model (Source: skills/image-prompting/SKILL.md:1-60). The Exa skill exposes four actions (search, answer, contents, find-similar) and example research workflows, costing $0.01 per call except contents which is $0.002 per URL (Source: skills/exa-research/SKILL.md:1-50, src/tools/exa.ts:14-21).

Community discussion around routing and trust has surfaced requirements that are relevant to the catalog design. A proposed feature (issue #14) asks for a free preflight trust check on each model's payTo Base wallet before routing — the architecture above already has the right hook point, since checkBudget runs before any settlement. Issue #7 advertises "41+ models with no API keys and 1M+ API calls/month on record," confirming that the catalog has grown beyond the earlier 30-model target and that agents are using the pay-per-call rail at production scale. The blockrun-mcp repo does not yet ship a SECURITY.md (issue #11), so vulnerabilities should be reported through GitHub's private advisory channel on the upstream repo.

See Also

• MCP Server & Protocol Handler
• Budget, Wallet & x402 Payments
• Adding a New Partner API
• Surf Endpoint Catalog
• Image Prompting Guide

Related Pages

Related topics: MCP Tools Catalog & Data Sources, Installation, Operations, Security & Extensibility

Skills, Endpoint Catalogs & Prompting

Overview

BlockRun MCP exposes a uniform set of payment-bearing tools to MCP clients, but the tool *descriptions* are intentionally minimal. The repository splits the heavy documentation load into three coordinated systems: skills (per-tool Markdown guides for the LLM agent), endpoint catalogs (path-based passthrough surfaces for APIs with many sub-endpoints), and prompting frameworks (reusable prompt templates that materially improve output quality). Together they keep the tool schemas short while moving all examples, gotchas, and reference material into files the agent can read on demand. Source: README.md.

Skills: Agent-Facing Documentation

What a skill is

A skill is a Markdown file under skills/<name>/SKILL.md with YAML front-matter. The front-matter declares a name, a human-readable description, and a list of trigger keywords that the MCP host uses to know when to load the skill. Source: skills/image-prompting/SKILL.md:1-18. Source: skills/search/SKILL.md:1-15.

The Markdown body is the actual instruction set: how to call the tool, what body shape to pass, a quick decision table, worked examples, common workflows, and a "gotchas" section. The skills are loaded lazily by the agent based on the trigger list, so a single MCP server can ship many skills without bloating every tool's description. Source: skills/exa-research/SKILL.md:1-25.

How skills map to tools

Each skill corresponds to one MCP tool registered in src/tools/. For example, the search skill documents the blockrun_search tool whose implementation lives at src/tools/search.ts:1-10. The skill explains the body shape (query, sources, max_results), the pricing curve, and the freshness trade-off, while the tool file enforces the budget check and forwards the request through the payment-bearing client. Source: skills/search/SKILL.md:19-45.

Endpoint Catalogs: Path-Based Passthrough

Several upstream APIs (Exa, Surf, Modal, etc.) have dozens of sub-endpoints that change over time. Rather than re-publishing each one as a new MCP tool on every change, BlockRun exposes one tool per upstream and routes by path/body. As of v0.14.1 the blockrun_exa tool is path-based: the agent passes the endpoint name and the request as a body. Source: skills/exa-research/SKILL.md:11-22.

The Surf catalog is the largest of these. It documents 84 endpoints organized by domain (On-chain, Prediction Markets, Project + DeFi, Search, Social, Token, Wallet, Web, Chat). Each entry in the catalog carries a path, a verb, a price tier (1, 2, or 3), and a notes column describing required parameters and known pitfalls. Source: skills/surf/SKILL.md:13-90. Some Surf highlights:

The Surf skill also surfaces cross-cutting gotchas: 56 of 84 endpoints require at least one param, missing params yield a 400 with no charge, and the Solana wallet works the same as the Base wallet for routing while settlement always lands in Surf's Base treasury. Source: skills/surf/SKILL.md:91-110.

Prompting Frameworks

Skills are not only about API shape. The image-prompting skill ships a full prompting framework because most image failures are prompt failures, not model failures. Source: skills/image-prompting/SKILL.md:21-30.

The 5-section framework

The framework is five short blocks separated by blank lines: SCENE, SUBJECT, DETAILS, USE CASE, and CONSTRAINTS. The first three describe the image; the fourth tells the model what kind of artifact to produce; the fifth bounds what must not drift (text, layout, faces, aspect ratio). The skill labels CONSTRAINTS as "where most mediocre prompts fail silently" and recommends restating it on every iterative edit to fight drift. Source: skills/image-prompting/SKILL.md:88-120.

Text and anti-slop rules

Two additional rulesets elevate quality. The text rules are: wrap literal text in quotes or ALL CAPS, specify font style/weight/color/placement, mark each line of copy with a role label (HERO:, SUB:, etc.), and spell difficult words letter-by-letter. The anti-slop rules replace vague praise with concrete visual facts ("overcast daylight, brushed aluminum, 50mm feel" instead of "stunning, epic, masterpiece") and name the lens and the light source. Source: skills/image-prompting/SKILL.md:42-87.

Pricing as a First-Class Concept

Cost is documented inside the skill, not just the tool. Search pricing is $0.025 per returned source, capped at 1–50, default 10 upstream, so a default call costs $0.25. The same constant is mirrored in the tool implementation so the budget guard can pre-check before a 402 round-trip. Source: src/tools/search.ts:18-32. Source: skills/search/SKILL.md:30-40. The image skill gives a per-model price ladder (gpt-image-2 ≈ $0.04, nano-banana ≈ $0.01, flux-1.1-pro $0.04–0.06) so the agent can pick a model on cost and quality jointly. Source: skills/image-prompting/SKILL.md:8-20.

See Also

• README.md for the full use-case catalog (chat, image, speech, phone, wallet delegation, cross-chain SQL).
• package.json for the runtime stack: @modelcontextprotocol/sdk ^1.0.0, @blockrun/llm ^2.11.0, viem ^2.21.0, zod ^4.3.5, Node >=18.

Related Pages

Related topics: Overview, Architecture & Payment Flow, MCP Tools Catalog & Data Sources, Skills, Endpoint Catalogs & Prompting

Installation, Operations, Security & Extensibility

This page is the operations reference for blockrun-mcp v0.21.1 (package.json:1-3). It covers how to install the MCP server, run it in development and production, manage its security posture, and extend it with new tools or skills.

Installation

blockrun-mcp is shipped as an npm package and exposes a single binary, blockrun-mcp, that MCP-compatible clients (Claude Code, Cursor, etc.) launch over stdio. The package targets Node.js ≥ 18 (package.json:60-62) and depends on the BlockRun SDK, the Anthropic SDK, viem for chain interactions, sharp for image processing, and the MCP SDK (package.json:46-58).

There are three supported installation paths documented in the project:

``bash claude mcp add blockrun -- npx -y @blockrun/mcp `` Source: README.md.

``bash git clone https://github.com/blockrunai/blockrun-mcp cd blockrun-mcp npm install npm run build claude mcp add blockrun-dev node /path/to/blockrun-mcp/dist/index.js `` Source: CONTRIBUTING.md.

1. Claude Code (one-shot add). The recommended path on the project page registers the server globally so it is available in every project:
2. Local source build for contributors. Clone, install, build, then point the client at the dev binary:
3. MCP stdio smoke test. A raw JSON-RPC handshake can be piped into the built server to verify it advertises 15 tools without needing a full MCP host. Source: CONTRIBUTING.md.

Operations

The npm scripts in package.json define the entire operational surface (package.json:25-31):

sharp is intentionally externalized in the build step so its native binary is not bundled, which is critical for cross-platform installs (package.json:25). In src/tools/search.ts the runtime cost of every call is pre-computed in JS from max_results and then asserted against a budget before payment is authorized, so operators can read the cost model directly from the tool source. Source: src/tools/search.ts:15-32.

Security

The repository has no SECURITY.md at the time of writing, and a community member has explicitly requested a private GitHub Security Advisory channel for responsible disclosure. Track this in issue #11 before reporting vulnerabilities. Source: community context — issue #11.

The following security controls are present in the codebase:

• Budget enforcement. Every tool calls checkBudget() before signing a payment and recordSpending() after a successful call. This is the primary guardrail against runaway spend on pay-per-use endpoints. Source: src/tools/search.ts:6-7.
• x402 USDC micropayments. The server is settlement-bound: each model call is paid in USDC on Base (or Solana) via the x402 protocol, and the server does not persist or proxy API keys for the upstream LLM providers. Source: package.json:1-3.
• Skill-trigger allowlist. Each skill declares a triggers: list in its YAML frontmatter that constrains when the agent should load it, reducing the attack surface for prompt-driven misuse. Source: skills/image-prompting/SKILL.md:1-12, skills/exa-research/SKILL.md:1-12.
• Counterparty trust (proposed). A community request (#14) proposes a free preflight check that scores any Base payTo address against a BLOCK/CAUTION/PROCEED rubric before the agent authorizes a payment. This is not yet implemented in source; operators handling untrusted model catalogs should monitor this issue.

Extensibility

The repository distinguishes two extension surfaces and applies the same naming convention to each (CONTRIBUTING.md):

flowchart LR
    A[New API to expose] --> B{Needs typed params,\nmulti-step logic, or\ncore LLM access?}
    B -- Yes --> T[Add MCP tool\nin src/tools/]
    B -- No --> S[Add Skill\nin skills/]
    T --> R[Rebuild dist/]
    S --> R
    R --> C[Restart MCP client]

• Add an MCP tool when the API needs Zod-validated enums, async polling, or holds core value (chat, image, search, voice). The reference implementation in src/tools/search.ts shows the canonical shape: a raw client, a per-call price estimate, a budget precheck, and a single McpServer.tool() registration. Source: src/tools/search.ts:1-32.
• Add a Skill when the API is simple, well-documented upstream, and primarily needs prompt framing. The existing skills in skills/ follow a Quick Decision Table → Endpoint examples → Gotchas → Reference structure that can be copied as a template. Source: skills/search/SKILL.md, skills/surf/SKILL.md, skills/exa-research/SKILL.md, skills/image-prompting/SKILL.md.

In all cases, run npm run typecheck && npm run build and re-run the stdio smoke test before opening a PR. Source: CONTRIBUTING.md.

See Also

• README.md — overview and quickstart.
• CONTRIBUTING.md — development workflow and tool/skill design rule.
• package.json — scripts, dependencies, engines.

Doramagic Pitfall Log

Found 11 structured pitfall item(s), including 1 high/blocking item(s). Top priority: Configuration risk - Configuration risk requires verification.

1. Configuration risk: Configuration risk requires verification

• Severity: high
• Finding: Project evidence flags a configuration risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: packet_text.keyword_scan | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

2. Identity risk: Identity risk requires verification

• Severity: medium
• Finding: Project evidence flags a identity risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: identity.distribution | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

3. Installation risk: Installation risk requires verification

• Severity: medium
• Finding: Project evidence flags a installation risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | cevd_810c75b756b9409aae5b8ed474881a07 | https://github.com/BlockRunAI/blockrun-mcp/issues/9

4. Configuration risk: Configuration risk requires verification

• Severity: medium
• Finding: Project evidence flags a configuration risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: capability.host_targets | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

5. Capability evidence risk: Capability evidence risk requires verification

• Severity: medium
• Finding: README/documentation is current enough for a first validation pass.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: capability.assumptions | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

6. Maintenance risk: Maintenance risk requires verification

• Severity: medium
• Finding: Project evidence flags a maintenance risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: evidence.maintainer_signals | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

7. Security or permission risk: Security or permission risk requires verification

• Severity: medium
• Finding: no_demo
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: downstream_validation.risk_items | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

8. Security or permission risk: Security or permission risk requires verification

• Severity: medium
• Finding: no_demo
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: risks.scoring_risks | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

9. Security or permission risk: Security or permission risk requires verification

• Severity: medium
• Finding: Project evidence flags a security or permission risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | cevd_c46a356b18224401b889a0208002efac | https://github.com/BlockRunAI/blockrun-mcp/issues/7

10. Maintenance risk: Maintenance risk requires verification

• Severity: low
• Finding: issue_or_pr_quality=unknown。
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: evidence.maintainer_signals | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

11. Maintenance risk: Maintenance risk requires verification

• Severity: low
• Finding: release_recency=unknown。
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: evidence.maintainer_signals | art_e028765b20a84b04af53db21f9414116 | https://github.com/punkpeye/awesome-mcp-servers#readme

Community Discussion Evidence

Doramagic exposes project-level community discussion separately from official documentation. Review these links before using blockrun-mcp with real data or production workflows.

• Multi-chain x402 facilitator for blockrun-mcp -- Algorand, Hedera, Stell - github / github_issue
• Paid ad slot on aibtc.news? - github / github_issue
• aibtc.news classifieds — BlockRun.ai listing, callable agent infra (3,00 - github / github_issue
• Configuration risk requires verification - GitHub / issue