Microsoft Agent Framework

# Fix list: Microsoft Agent Framework

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/microsoft-agent-framework, the October 2026 research run, assessed 8 October 2026. Grade A, 82.2 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Microsoft Agent Framework: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: Read with the self-hosted rule. No payment protocol (0). The package is free under MIT, and Foundry hosted agents, the managed option, has per-unit prices in Azure's public retail price API ($0.0994 a vCPU-hour and $0.0118 an hour for memory in East US). The pricing web page showed no figures to our reader (20). The packages install with no card (20) and no account, and run local models through the Ollama and Foundry Local connectors (20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 2. Reliability, 78 out of 100, up to 4.4 more on the total

Why it scored 78: Read with the local-software lines. Official packages on PyPI (Requires-Python >=3.10) and NuGet (net8.0 to net10.0, netstandard2.0, net472) (20). The .NET build and test workflow passed on 11 of the 12 most recent pushes to main on 8 October 2026, with one cancelled, and the Python merge tests passed on each nightly run from 1 to 8 October (25). 575 open issues and 201 open pull requests, with 260 issues closed in the 30 days to 8 October and 14 open issues labelled bug (10). The Python changelog follows Keep a Changelog and states semantic versioning, and marks breaking changes, but they shipped in minor releases of released packages, such as agent-framework-core in 1.20.0 on 2 October 2026 (8). Python 1.21.0 and .NET 1.24.0, classed 5 - Production/Stable on PyPI (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 3. Agent ergonomics, 87 out of 100, up to 2.1 more on the total

Why it scored 87: The documented Python agent with one MCP server is about 12 lines with `MCPStdioTool` from the core package, and `allowed_tools`, `load_tools` and progressive disclosure limit what reaches the model (25). The function loop stops at 40 iterations by default with a `max_function_calls` option, and compaction has its own docs page (20). A typed exception hierarchy, and the loop stops after three consecutive tool errors by default. We didn't find MCP failures documented (16). Workflows resume from checkpoints and sessions serialise, but we found no retry setting for model calls in the core client (14). An agent needs a client and instructions. Python and .NET are stable, Go is in preview, and the quickstarts assume a Microsoft Foundry project (12).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 4. Transparency & trust, 83 out of 100, up to 1.5 more on the total

Made of editorial 86, provenance 80.

Why it scored 83: MIT (30). The core README says what the User-Agent and feature token are, and a design spec in the repository says the token goes only to Foundry and Azure OpenAI pipelines and not to third-party providers. That spec is marked proposed, and we found no statement of how long Microsoft keeps the data and no privacy document for the framework (22). Package lifecycle states are published per package, experimental APIs raise warnings, breaking changes are dated in the Python changelog and there are upgrade guides, but we found no written support window (14). Telemetry is disclosed with two opt-out variables (20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Status page: not found (0 of 10)
- security.txt: published but past its Expires date (5 of 10)

## 5. Security & auth, 93 out of 100, up to 1.2 more on the total

Why it scored 93: Framework reading, so SOC 2 isn't scored. A package and version User-Agent goes out on client requests by default, with a feature-usage token on Foundry and Azure OpenAI paths. Both are documented with opt-out variables, and the observability module says the framework contacts no external system itself, with message content in traces off by default (25). `approval_mode` on function and MCP tools, an MCP tool allow-list, and Hyperlight and Monty sandboxes for model-written code in beta. Tools run without approval unless set (20). An Agent Safety page treats model-supplied arguments as untrusted, and FIDES labels content and enforces policy before sensitive tools run, though it is experimental and Python only (15). OpenTelemetry traces, metrics and logs to any exporter (15). SECURITY.md routes reports to MSRC with a bounty programme and coordinated disclosure, CodeQL runs in CI and three advisories were published with fixed versions on 7 October 2026. Microsoft's security.txt expired on 23 September 2026 (18).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 6. Schema & documentation, 95 out of 100, up to 0.8 more on the total

Why it scored 95: Typed in both languages, with py.typed markers and API references for Python and .NET on Microsoft Learn (25). Learn returns each docs page as Markdown when asked with an Accept header. No llms.txt was found (10). The overview has a table for when to use an agent or a workflow, and says to write a function instead where one would do (20). Tools are Python functions typed with Annotated and Pydantic Field, or AIFunction in .NET (15). Samples for every area and a typed exception hierarchy in the source, but we found no exception reference in the docs pages we read (12). A dated Python changelog and four upgrade guides. No changelog file was found for .NET (13).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Maintenance & community, 91 out of 100, up to 0.8 more on the total

Why it scored 91: Python 1.21.0 on 2026-10-08 and .NET 1.24.0 on 2026-10-07 (30). 23 release tags in the 90 days to 8 October, 12 Python and 11 .NET (20). 575 open issues and 201 open pull requests, 260 issues closed in 30 days, and triage and labelling workflows run on new issues. We couldn't see reply times (17). Python and .NET packages are current, and Go is a prerelease at v0.1.0 from 1 September 2026 (15). CI, CodeQL and Dependabot run, though two of four Dependabot update runs failed on 8 October (9).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 2026-10-07. High-severity advisory GHSA-9f26-7mqc-9vc4 (CVSS 7.5). Microsoft.Agents.AI.Workflows.Declarative before 1.24.0 could forward values marked sensitive, including environment-scoped secrets, to an external MCP server from an InvokeMcpTool action. Fixed in 1.24.0 and published the same day, so 2 points. Two medium advisories published the same day (GHSA-q7g7-7c7x-frcq, approval not bound to transport headers, and GHSA-r2mq-wgc9-vgh3, Foundry toolbox bearer token sent to an unintended origin) weren't deducted. https://github.com/microsoft/agent-framework/security/advisories/GHSA-9f26-7mqc-9vc4

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: prices on the Foundry Agent Service pricing page. Our reader saw placeholders where the figures load, so the hosted agent prices are from Azure's retail price API for East US, and the unit of the memory meter is given there only as 1 Hour
- unchecked: reply times on issues. We counted open and closed issues through the GitHub API and did not read threads
- unchecked: a retry setting for model calls. None was found in the core client source, and the provider SDKs may retry on their own
- unchecked: the .NET release notes. No changelog file is in the repository and we did not read the GitHub release pages
- unchecked: the Python and .NET API reference pages. Both URLs answered with a redirect and we did not read their content
- No terms or privacy document specific to the framework was found, so provenance.terms and provenance.privacy are left out
- The feature-usage telemetry spec is marked proposed while the code ships the token. We found no statement of how long Microsoft keeps User-Agent data
- The support FAQ says .NET and Python are the supported languages, while the overview adds Go in public preview

## Weaknesses

- Changes marked [BREAKING] shipped in minor releases of released packages, including agent-framework-core in 1.20.0 on 2 October 2026
- A package and version User-Agent goes out on client requests by default, with a feature-usage token on Foundry and Azure OpenAI paths. Two environment variables turn them off
- Three advisories were published on 7 October 2026 against .NET packages, one high (CVSS 7.5), all with fixed versions
- Tools run without approval unless approval is set on each tool, per the safety page
- 575 open issues and 201 open pull requests on 8 October 2026, and 26 of 44 Python packages are alpha or beta

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Pin exact versions. Minor releases have carried [BREAKING] changes, so read python/CHANGELOG.md or the upgrade guides before moving up
- Set `approval_mode="always_require"` on tools and MCP servers that write. The default is `never_require`
- Pass `allowed_tools` to `MCPStdioTool` or `MCPStreamableHTTPTool` so only the named MCP tools reach the model
- Set `AGENT_FRAMEWORK_USER_AGENT_DISABLED=true` to drop the framework's User-Agent and feature token, or `AGENT_FRAMEWORK_FEATURE_MASK_DISABLED=true` for the token alone
- On .NET, use Microsoft.Agents.AI.Workflows.Declarative 1.24.0 or later. Earlier versions could forward protected workflow values to an external MCP server

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.