The Intelligence Radar MCP server
OfficialServer Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
The Intelligence Radar (H2 2026, Volume 1: Room to act), is a community-driven radar for technical leaders, managed by Brightbeam. It holds 103 entries in 9 dimensions. Each entry has an intelligence label, for readiness under the conditions it states (Use, Trial, Evaluate, Wait, Prepare), and an enterprise label, for what an organisation should do now (Revisit, Test, Plan for, Watch). A label holds under the conditions its entry states, so read the entry before applying it to a situation. Evidence runs to 28 October 2026. Start with search_radar or check_plan; use get_entry for the judgement, the test to run first, the failure to watch for and the sources. For financial services or insurance or pharma and biotech or medtech or manufacturing or public services, get_sector gives the sector lens: what binds there, where to start and the entries placed differently; search_radar, check_plan and show_radar take a sector too. When you use the Radar in an answer, name each entry by ID and title (for example IR-028 MCP tool integration), give both labels and link its page. Cite a figure by its source number and link. To contribute to The Intelligence Radar, start with how_to_contribute; show the person the preview from prepare_contribution and submit only with their explicit agreement.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| prompts | {
"listChanged": true
} |
| resources | {
"listChanged": true
} |
| completions | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| about_radarA | What The Intelligence Radar is, this edition, how many entries carry each label, the nine dimensions and how to use this server. Call it first if you have not used the Radar before. |
| search_radarA | Find entries by topic, technology, practice or keyword, optionally filtered by dimension, intelligence label or enterprise label. Returns each entry's ID, title, dimension, both labels, a one-sentence judgement and its web page. With filters and no query, lists the matching entries in order. |
| get_entryB | The full entry: the judgement and the conditions under which it holds, where it fits, the test to run first, the failure to watch for, examples, the owner, related enterprise judgements and the numbered sources with links. |
| list_dimensionsA | The nine dimensions, grouped on three lenses (capability, context and control), each with its question and its entries by intelligence label. |
| get_dimensionB | One dimension: what it covers, why its entries sit together, what its placements imply for an organisation, and its entries by intelligence label. |
| check_planA | Match a plan, roadmap, architecture, policy, business case or list of initiatives against The Intelligence Radar. Returns the entries the text touches, with both labels, the conditions to check, the test to run first and the failure to watch for. Entries at Wait or Prepare are flagged, because the plan may assume readiness the Radar does not support, and dimensions the text does not touch are listed. |
| list_judgementsA | This edition's enterprise judgements: for each, the enterprise label, the condition, the evidence, the action, the trade-off, when to review it and the related entries. |
| get_sourcesB | The numbered sources the Radar cites as [n], with title, links, kind and how that kind of source is read. |
| get_sectorA | How the Radar reads for one sector: what binds there (the rules and their dates), the entries to start with, the entries placed differently there and why, what to hold back, and the entries that matter most. Without a sector, lists the sectors the lens covers. |
| show_radarA | Show the interactive radar in the conversation, optionally filtered by dimension or label, or with entries highlighted (for example the entries a search or plan check returned). People can select a point to read the entry and ask about it. Hosts without interactive views receive the same selection as text. |
| how_to_contributeA | How to contribute to The Intelligence Radar from this conversation: what can be contributed, what reviewers need, how review and publication work, and whether this assistant is connected to a GitHub account. Call it before helping someone contribute or review. |
| sign_in_with_githubA | Connect this assistant to the person's GitHub account, so that contributions and reviews go to GitHub from their account. The first call returns a short code and an address: ask the person to open the address and enter the code, then call this again to finish. The sign-in is kept on this computer. With sign_out true, it removes the sign-in. |
| prepare_contributionA | Check a contribution to The Intelligence Radar against this edition and prepare it for submission: a review, a correction, evidence, a label change or a new entry. It sends nothing. It returns the exact preview, any problems to fix, the payload and a draft_id. Show the person the preview and get their explicit agreement before calling submit_contribution. For a correction, copy the exact current words from get_entry into find; each must occur once in its section. Ask the person for anything missing, such as the interests they declare and how they want to be credited; never invent it. |
| submit_contributionA | Submit a contribution prepared with prepare_contribution in this conversation, by its draft_id. Call it only after showing the person the preview and getting their explicit agreement to submit it. Connected to GitHub, it opens the contribution as an issue from the person's account and returns the review desk's answer; otherwise it returns a pre-filled GitHub issue link for the person to open, check and submit. |
| my_contributionsA | The status of your contributions to The Intelligence Radar, from GitHub: all of them, or one in full with the reviewers' decisions and reasons. |
| revise_contributionA | Send a revised version of your contribution while it is in review or after the reviewers asked for changes. First prepare the revision with prepare_contribution (the same kind), show the person the preview and get their explicit agreement; then call this with the contribution's id and the new draft_id. It updates the GitHub issue; approvals of earlier versions no longer count. |
| withdraw_contributionA | Withdraw your contribution before a final decision. Call it only when the person has said so. It comments /withdraw on the GitHub issue, which closes it. |
| apply_to_contributeA | A pre-filled link to the form for joining the list of people of The Intelligence Radar as a contributor or specialist reviewer. The application is a public GitHub issue that an editor reviews. Fill in only what the person gives. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| brief_dimension | A one-page brief on one dimension of the Radar for your leadership team or board. |
| plan_trial | Turn an entry's test into a 30-day trial plan with measures, thresholds and a decision at the end. |
| review_roadmap | Check an AI roadmap, plan or architecture against The Intelligence Radar: what is well placed, what is at risk and what is missing. |
| board_questions | Questions a board or executive committee should ask about an entry or a dimension, with what a good answer looks like. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| The Intelligence Radar view | The interactive radar: every entry by dimension and ring, with filters and the entry beside it. |
| about | The edition, the labels, the dimensions and how to use this server |
| labels | Both label scales, how to read them together, and the evidence labels |
| full-text | Every page, figure description, entry and source of the edition in one Markdown file |
| 1 Models and inference | One dimension with what its placements imply and its entries |
| 2 Knowledge, context and expertise | One dimension with what its placements imply and its entries |
| 3 Agents and workflow | One dimension with what its placements imply and its entries |
| 4 Assurance, security and governance | One dimension with what its placements imply and its entries |
| 5 Evaluation and learning | One dimension with what its placements imply and its entries |
| 6 Prediction, science and physical AI | One dimension with what its placements imply and its entries |
| 7 Work, products and people | One dimension with what its placements imply and its entries |
| 8 Operations, economics and sovereignty | One dimension with what its placements imply and its entries |
| 9 Compute and hardware | One dimension with what its placements imply and its entries |
| IR-001 Frontier model selection | One entry in full, with its sources |
| IR-002 Small language models | One entry in full, with its sources |
| IR-003 Open-weight self-hosting | One entry in full, with its sources |
| IR-004 Multimodal document understanding | One entry in full, with its sources |
| IR-005 Speech and real-time voice | One entry in full, with its sources |
| IR-006 Long context versus retrieval | One entry in full, with its sources |
| IR-007 Structured outputs and typed contracts | One entry in full, with its sources |
| IR-008 Reasoning and inference-time budgets | One entry in full, with its sources |
| IR-009 Model routing and cascades | One entry in full, with its sources |
| IR-010 Quantisation and inference optimisation | One entry in full, with its sources |
| IR-011 Caching and batch inference | One entry in full, with its sources |
| IR-012 Speculative decoding and distributed serving | One entry in full, with its sources |
| IR-013 Permission-aware retrieval | One entry in full, with its sources |
| IR-014 Hybrid keyword and vector search | One entry in full, with its sources |
| IR-015 Reranking and retrieval evaluation | One entry in full, with its sources |
| IR-016 Document parsing and OCR | One entry in full, with its sources |
| IR-017 Knowledge graphs and GraphRAG | One entry in full, with its sources |
| IR-018 Context assembly and compression | One entry in full, with its sources |
| IR-019 Agent memory with owners and expiry | One entry in full, with its sources |
| IR-020 Expertise and tacit knowledge capture | One entry in full, with its sources |
| IR-021 Evidence provenance and citations | One entry in full, with its sources |
| IR-022 Data contracts and semantic definitions | One entry in full, with its sources |
| IR-023 Synthetic data for gaps and testing | One entry in full, with its sources |
| IR-024 Freshness, correction and deletion propagation | One entry in full, with its sources |
| IR-025 Bounded single-agent workflows | One entry in full, with its sources |
| IR-026 Deterministic workflow with model steps | One entry in full, with its sources |
| IR-027 Durable execution and idempotency | One entry in full, with its sources |
| IR-028 MCP tool integration | One entry in full, with its sources |
| IR-029 Agent-to-agent interoperability | One entry in full, with its sources |
| IR-030 Multi-agent orchestration | One entry in full, with its sources |
| IR-031 Computer-use and browser agents | One entry in full, with its sources |
| IR-032 Human approval and takeover | One entry in full, with its sources |
| IR-033 Reusable agent skills and prompt packages | One entry in full, with its sources |
| IR-034 Agent sandboxes and egress control | One entry in full, with its sources |
| IR-035 Coding agents across delivery | One entry in full, with its sources |
| IR-036 Unattended open-ended autonomy | One entry in full, with its sources |
| IR-037 Agent identities and least privilege | One entry in full, with its sources |
| IR-038 Prompt-injection resilience | One entry in full, with its sources |
| IR-039 Secrets, data loss and output validation | One entry in full, with its sources |
| IR-040 Independent test and grading environments | One entry in full, with its sources |
| IR-041 Red-teaming and adversarial rehearsal | One entry in full, with its sources |
| IR-042 AI supply-chain verification | One entry in full, with its sources |
| IR-043 Structured oversight records | One entry in full, with its sources |
| IR-044 Privacy engineering and minimisation | One entry in full, with its sources |
| IR-045 Incident containment and kill switches | One entry in full, with its sources |
| IR-046 Formal verification and proof assistants | One entry in full, with its sources |
| IR-047 Conformance evidence over certification claims | One entry in full, with its sources |
| IR-048 High-stakes autonomous decisions | One entry in full, with its sources |
| IR-049 Owned task examinations | One entry in full, with its sources |
| IR-050 Model-and-harness evaluation | One entry in full, with its sources |
| IR-051 Judge models with human calibration | One entry in full, with its sources |
| IR-052 Accepted work and full-cost measurement | One entry in full, with its sources |
| IR-053 Drift, slices and uncertainty | One entry in full, with its sources |
| IR-054 Shadow and canary deployment | One entry in full, with its sources |
| IR-055 Fine-tuning and low-rank adapters | One entry in full, with its sources |
| IR-056 Distillation and specialist classifiers | One entry in full, with its sources |
| IR-057 Reinforcement learning and post-training | One entry in full, with its sources |
| IR-058 Governed learning releases | One entry in full, with its sources |
| IR-059 Harness ablation and simplification | One entry in full, with its sources |
| IR-060 Unreviewed self-modifying production agents | One entry in full, with its sources |
| IR-061 Classical tabular ML baselines | One entry in full, with its sources |
| IR-062 Tabular foundation models | One entry in full, with its sources |
| IR-063 Time-series foundation models | One entry in full, with its sources |
| IR-064 Causal inference and experiments | One entry in full, with its sources |
| IR-065 Optimisation and constraint solvers | One entry in full, with its sources |
| IR-066 Computer vision for inspection | One entry in full, with its sources |
| IR-067 Graph analytics and relational prediction | One entry in full, with its sources |
| IR-068 Simulation and digital twins | One entry in full, with its sources |
| IR-069 Scientific discovery with AI | One entry in full, with its sources |
| IR-070 Edge inference and on-device AI | One entry in full, with its sources |
| IR-071 Robotics and vision-language-action systems | One entry in full, with its sources |
| IR-072 Unconstrained physical autonomy | One entry in full, with its sources |
| IR-073 Document drafting and analysis assistants | One entry in full, with its sources |
| IR-074 Customer-service copilots | One entry in full, with its sources |
| IR-075 Contact-centre automation | One entry in full, with its sources |
| IR-076 Enterprise search and knowledge assistance | One entry in full, with its sources |
| IR-077 Process mining before automation | One entry in full, with its sources |
| IR-078 Expert workflows and decision support | One entry in full, with its sources |
| IR-079 Creative and synthetic media workflows | One entry in full, with its sources |
| IR-080 Human learning and apprenticeship | One entry in full, with its sources |
| IR-081 AI literacy and supported adoption | One entry in full, with its sources |
| IR-082 Accessible and multilingual AI services | One entry in full, with its sources |
| IR-083 Agent commerce and delegated purchasing | One entry in full, with its sources |
| IR-084 Organisation-wide productivity claims | One entry in full, with its sources |
| IR-085 AI service and dependency inventory | One entry in full, with its sources |
| IR-086 Observability with purposeful traces | One entry in full, with its sources |
| IR-087 Evaluation-gated releases and model retirement | One entry in full, with its sources |
| IR-088 Cost per accepted task and FinOps | One entry in full, with its sources |
| IR-089 Rate limits, queues and spend controls | One entry in full, with its sources |
| IR-090 Resilience and supplier exit | One entry in full, with its sources |
| IR-091 Sovereignty beyond data residency | One entry in full, with its sources |
| IR-092 EU and national regulatory readiness | One entry in full, with its sources |
| IR-093 Validated change in regulated operations | One entry in full, with its sources |
| IR-094 Compute capacity, energy and location | One entry in full, with its sources |
| IR-095 National outcome evidence and sector benchmarks | One entry in full, with its sources |
| IR-096 Continuous operation of AI systems | One entry in full, with its sources |
| IR-097 Accelerators beyond a single vendor | One entry in full, with its sources |
| IR-098 Inference-specialised hardware | One entry in full, with its sources |
| IR-099 Owning, renting and reserving GPU capacity | One entry in full, with its sources |
| IR-100 GPU cloud and capacity-contract risk | One entry in full, with its sources |
| IR-101 Memory supply and hardware prices | One entry in full, with its sources |
| IR-102 AI PCs and NPUs in the fleet | One entry in full, with its sources |
| IR-103 Risk-tiered governance with named owners | One entry in full, with its sources |
TDQS
Scored across 18 tools
Most tools target a clearly distinct resource plus action: read-side (search_radar, get_entry, get_dimension, list_dimensions, list_judgements, get_sources, get_sector, check_plan) and contribution-side (prepare/submit/revise/withdraw/my_contributions) are cleanly separated, and prepare vs submit is explicitly spelled out. The main soft spots are search_radar vs show_radar (both can surface a filtered set of entries) and about_radar vs how_to_contribute, which an agent could plausibly mix up on a first call.
Names are uniformly snake_case and overwhelmingly follow verb_noun (submit_contribution, get_entry, list_dimensions, check_plan, sign_in_with_github). A few break the verb pattern (my_contributions, about_radar, how_to_contribute), but they are readable and internally sensible rather than chaotic.
18 tools is on the heavy side, but the server genuinely spans three surfaces: content discovery/reading, plan and radar analysis, and a full contribution lifecycle plus GitHub auth. Nearly every tool earns its place, though the read side (search/get/list/show/dimension/sector/sources) is dense enough that some consolidation could be considered.
The read surface is thorough (search, entry, dimension, judgements, sources, sector, plan check, interactive view) and the contribution lifecycle covers prepare, submit, revise, withdraw, status, sign-in/out and applying as a contributor. Minor gaps: how_to_contribute references reviewing contributions, yet no reviewer-facing tool exists in the set, and there is no direct 'browse by label' entry point beyond search filters.